CN118172676B - A method for detecting farmland pests based on quantum deep learning - Google Patents

Abstract

The invention discloses a farmland pest detection method based on quantum deep learning, which relates to the technical field of quantum deep learning, and comprises the steps of collecting detected pest images, and converting the collected detected pest images into quantum images by utilizing NEQR models; constructing an improved deep convolutional neural network to classify and identify pest images; optimizing the image by using a novel enhanced quantum representation model; extracting data features of pest images, classifying pests with the same data features, and establishing a farmland pest classification database; and collecting images of farmland with pests, performing image edge detection by using pixel gray scale association, determining the types and the quantity of pests in each area of the farmland according to a farmland pest classification database, and intensively spraying corresponding pesticides. The method has the advantages of higher image accuracy, high operation precision and high operation efficiency.

Description

A method for detecting farmland pests based on quantum deep learning

Technical Field

The present invention relates to the field of quantum deep learning technology, and in particular to a method for detecting farmland pests based on quantum deep learning.

Background Art

Image recognition methods for detecting pests were first used in the United States in the 1860s, using machine vision technology to detect adult grain borers offline. This technology can identify 90% of the main stored-grain pests, but factors such as damaged grains, grass seeds, and the morphology of the pests still have a great impact on the correct identification of pests.

At present, there are two main forms of farmland pest detection: one is the traditional target detection algorithm, and the other is the algorithm detection based on deep learning. The traditional target detection process is roughly divided into candidate region selection, feature extraction and category judgment. Although many operators and generalization methods are designed in the traditional algorithm for selecting candidate regions, feature extraction and classification, the improvement of the model is not obvious. There are still several major problems: (1) The generated candidate boxes have a lot of redundancy, which increases the amount of calculation; (2) The sliding window cannot adapt to the transformation of the target scale; (3) The feature operator only extracts low-dimensional visual features and loses the semantic relationship between the target and the environment in the image. The development and application of deep learning technology has greatly improved the accuracy and efficiency of target detection, but there are still problems such as large amount of calculation, high hardware cost and complex model design. It can be seen that how to improve the detection efficiency and how to automatically and accurately detect farmland pests are currently urgent problems to be solved.

For convolutional neural networks, a large amount of data sets are needed to train these convolution kernels. Since the training data has different scales, they need to be normalized to the same scale for training. Therefore, convolutional neural networks have the characteristics of long training time, multiple and complicated training models.

Summary of the invention

In order to overcome the above problems existing in the prior art, the present invention proposes a method for detecting farmland pests based on quantum deep learning.

The technical solution adopted by the present invention to solve the technical problem is: a method for detecting farmland pests based on quantum deep learning, which specifically includes the following steps:

Step 1, collecting images of detected pests, and converting the collected images of detected pests into quantum images using a novel enhanced quantum representation model;

Step 2, preprocessing the pest image obtained in step 1, and storing the image position information in the quantum system;

Step 3: Build an improved deep convolutional neural network to classify and identify pest images. If they are pests, input the pest image into the image segmentation network to reconstruct the pest image.

Step 4, optimizing the image obtained in step 3 using the new enhanced quantum representation model;

Step 5, extracting data features of pest images, and classifying pests with the same data features to establish a farmland pest classification database;

Step 6: Collect images of farmland with pests, use pixel grayscale association to detect image edges, determine the types and numbers of pests in each area of the farmland based on the farmland pest classification database obtained in step 5, and spray corresponding pesticides in a centralized manner.

In the above-mentioned method for detecting farmland pests based on quantum deep learning, the expression of the novel enhanced quantum representation model in step 1 is:

;

in, Represents a grayscale image; Represents binary grayscale information that can be used for encoding; n represents the length of the quantum bit; Represents a tensor product calculation in quantum computing.

In the above-mentioned method for detecting farmland pests based on quantum deep learning, the deep convolutional neural network in step 3 is an optimized YOLOv5 model, and the optimized YOLOv5 model includes an input end, a support layer, a fusion feature layer and an output layer. The input end uses a mosaic algorithm to achieve data enhancement; the support layer uses a focusing structure to cut the input image, cuts the image into four parts, and performs a convolution operation after splicing through longitudinal channels; the path aggregation network in the fusion feature layer is responsible for feature fusion, and adds an information flow path based on the feature extractor; during the model training process, the stochastic gradient descent algorithm is used to update the model parameters.

In the above-mentioned method for detecting farmland pests based on quantum deep learning, the step 3 specifically includes: using the preprocessed image obtained in step 2 as the input of the optimized YOLOv5 model, and representing each frame of the input image as x;

The output of the optimized YOLOv5 model includes the coordinate information of the target pest location and the category probability; for each input image x, the output is ;

Among them, m represents the number of targets detected in the image. Represents the prediction result of the i-th target; each target prediction result Bounding box coordinates including the target location And the corresponding class probability , the target position coordinate formula is:

;

in, represents the coordinates of the upper left corner of the bounding box, , Represent the width and height of the bounding box respectively; the category probability formula is as follows: ;

Where c represents the number of target pest categories, It represents the probability that the i-th target belongs to the j-th category.

The above-mentioned method for detecting farmland pests based on quantum deep learning, wherein step 4 specifically includes:

Step 4.1, convert the digital image into a quantum image: the image size is , the grayscale range is , then the NEQR model is expressed as follows:

;

In the formula, is the image quantum state;

;

image It consists of two parts, namely and Respectively represent the grayscale information and position information of the stored image; position information It consists of vertical coordinates and horizontal coordinates, and the specific expression is:

;

Among them, the first n quantum bits store the vertical coordinate information of the image, and the last n quantum bits store the horizontal coordinate information of the image;

The NEQR model representation of a binary image is:

;

In the formula, Indicates the binary information of the stored image; Indicates the location information of the stored image;

Step 4.2, use quantum logic gates to transform the storage state: the initial state of the quantum system is:

;

Applying H-gate transformation, each qubit representing position information is transformed by H-gate transformation, which can generate The superposition system of states is:

;

According to the pixel information of the actual image and the mapping relationship between coordinates and grayscale information, the initial state sequence representing the grayscale information in the quantum image is converted into the corresponding M state, realizing the unique mapping process between the position information and the grayscale information in the quantum image. The specific expression is:

;

Where n represents the length of the quantum bit; It represents all the ground states in the quantum grayscale image system; Represents tensor product calculation in quantum computing; Represents binary grayscale information that can be used for encoding; Represents a grayscale image;

Step 4.3, perform quantum measurement on the image quantum state: position information measurement operator for:

;

In the formula, represents the direct product of q identity matrices;

The grayscale value from the quantum state is recovered using projection measurement. The measurement operator M of the grayscale information is:

;

in, and Both represent the location information of the stored image.

In the above-mentioned method for detecting farmland pests based on quantum deep learning, the step 6 of using pixel grayscale association to perform image edge detection specifically includes:

Step 6.1, convert the image into a quantum image:

Digital image after grayscale processing Indicates at point The gray value at , ;like Represents pixel The probability that the gray value is 1, is the probability that the gray value is 0, then the image The available quantum bits can be represented as:

;

Step 6.2, describe the grayscale transition of the neighboring pixels: , and The gray values of three adjacent pixels are , and , according to the principle of quantum state superposition, they constitute a neighborhood grayscale correlation system, expressed as:

;

In the image quantum correlation system, the state vector is composed of three binary bits. From left to right, the state vector with a jump of 01 is a positive state vector, and the state vector with a jump of 10 is a negative state vector.

According to the difference operation between the probability of positive and negative state vectors and the probability of their corresponding anti-state vectors, we have

;

in, is the superposition vector The probability amplitudes corresponding to the previous coefficients respectively;

Step 6.3, construct edge detection template using average formula:

The grayscale transition of the image is described by the sum of the detection components in the horizontal and vertical directions, that is, the edge grayscale: ;

in, Indicates the image transition, that is, the edge grayscale, represents the detection component in the horizontal direction, Represents the detection component in the vertical direction;

Step 6.4, edge detection threshold setting:

The triple mean of the edge grayscale is used as the edge detection threshold, that is:

;

in, represents the threshold for edge detection, Represents the edge grayscale mean;

Extract binary edge image function ; If the edge grayscale mean is greater than the edge detection threshold, the binary edge image function is 0, otherwise the binary edge image edge function is 1.

The beneficial effects of the present invention are: (1) Quantum algorithm optimization for images: Based on convolutional neural networks, the NEQR quantum image model is applied to the construction and optimization of pest and disease detection images. At the same time, the quantum threshold denoising algorithm is used to effectively remove noise in the image, improve the clarity and quality of the image, and the gray correlation algorithm is used to detect image edges and smooth the image to make the image more accurate.

(2) Quantum algorithms improve image processing efficiency. Using quantum algorithms to detect edges based on pixel grayscale associations can improve the accuracy of image edge recognition, thereby more accurately capturing edge information in the image. Using convolutional neural networks to detect and recognize images can greatly speed up image processing. At the same time, the YOLOv5 algorithm can be used to make up for the shortcomings of deep learning technology in image processing, increase computational accuracy, improve computational efficiency, and use the NEQR model to make up for the shortcomings of convolutional neural networks.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention is further described below in conjunction with the accompanying drawings and embodiments.

FIG1 is a schematic diagram of the technical route of the present invention;

FIG. 2 is a diagram of an embodiment of the present invention. NEQR image and quantum state diagram;

FIG3 is a flow chart of classifying and identifying pest images in an embodiment of the present invention;

FIG4 is a diagram of a YOLOv5 model network architecture according to an embodiment of the present invention.

DETAILED DESCRIPTION

In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

As shown in Figure 1, this embodiment discloses a method for detecting farmland pests based on quantum deep learning. This embodiment applies a novel enhanced quantum representation (NEQR) quantum image to optimize the pest detection image, and uses a target detection algorithm (YOLOv5) to improve the accuracy of the image. Since the color and position of each image pixel are represented by classical numbers, this embodiment uses a novel enhanced quantum representation (NEQR) encoding method to use quantum states and their coefficients to represent the pixel position and color information in the image.

The main goal of the farmland pest image dataset is to assist researchers in quickly and accurately identifying the number of pest species in farmland. The machine learning algorithm can automatically identify and classify farmland pest images by training and testing the dataset. The farmland pest dataset constructed in this embodiment selects 10 normal farmland crop images and 20 pest farmland images.

The specific implementation method of this embodiment is shown in FIG1 , and specifically comprises: Step 1, collecting images of detected pests, and converting the collected images of detected pests into quantum images using a novel enhanced quantum representation model.

In order to determine the specific characteristic values of pests, it is necessary to store their position information and grayscale information, and the NEQR expression of the quantum grayscale image is:

;

in, Represents a grayscale image; Represents binary grayscale information that can be used for encoding; n represents the length of the quantum bit; Represents the tensor product calculation in quantum computing. M is the decimal representation of the grayscale value. When extended to the quantum image system, it needs to be transformed into a binary quantum bit sequence. It is worth noting that the number of quantum bits representing the grayscale information is related to the grayscale dynamic range of the prepared image. j is the decimal representation of the position information. Similarly, the position information of the image needs to be expanded into a binary quantum bit sequence. Where n represents the length of the quantum bit, and

;

It represents all the ground states in the quantum grayscale image system; It is the tensor product symbol in quantum computing and an important operation method for realizing operations between quantum logic gates.

Step 2: pre-process the pest image obtained in step 1 and store the image position information in the quantum system.

After the pest image data is calibrated, it is necessary to use a threshold denoising algorithm to denoise the data, and then enhance the pest image in order to make the characteristic values of the pest more obvious, so as to facilitate image recognition and feature collection. The display effect of pest features can be enhanced by methods such as contrast enhancement. In digital image processing, each pixel contains position information and grayscale information, and the pixel content can be accessed based on the position information in the pest image. Therefore, in quantum images, in order to store the position information in the quantum system, the position information needs to be split into a quantum sequence expression in binary form, that is,

;

In the formula, Represents information in the y direction; Represents information in the x direction.

From the above, we can see that a For a grayscale image of size 1, the grayscale values of the four pixels are 0, 100, 200, and 255 respectively. The corresponding NEQR expression can be expressed as:

;

In the formula, M ₀ , M ₁ , M ₂ , and M ₃ are expressed in binary sequence, representing the grayscale information in the quantum image, and the grayscale information corresponding to each pixel position is a unique mapping relationship, that is, one position information corresponds to a uniquely determined grayscale information, as shown in FIG2 .

Grayscale images only contain brightness information, not color information, and their brightness changes continuously from dark to bright. Grayscale is a logarithmic division of white and black into several levels, and its range is 0-255. Compared with the original color image, the grayscale image does not contain color information, so the amount of information contained in the grayscale image is greatly reduced, and the amount of image processing calculation is also greatly reduced, which is convenient for subsequent calculations.

Step 3: Build an improved deep convolutional neural network to classify and identify pest images. If they are pests, input the pest image into the image segmentation network to reconstruct the pest image, as shown in Figure 3.

The deep convolutional neural network model is trained using the public dataset, ImageNet, and the model parameters are continuously adjusted through quantum algorithms. During the training process, the model gradually learns the features of the image and converts them into feature vectors. These feature vectors can be used for subsequent image processing and applications.

The deep convolutional neural network model in this embodiment is an optimized YOLOv5 model.

Compared with other methods, the YOLOv5 algorithm has improved detection accuracy and speed, and introduced new design ideas and strategies. The algorithm adopts a single-stage detection, which can detect targets in the entire image at one time and realize real-time detection tasks. The design without anchor boxes directly predicts the position and size of the bounding box, which improves the adaptability and generalization ability of the model.

The optimized YOLOv5 model network architecture is shown in Figure 4. The target detection algorithm (YOLOv5) can be divided into the input end, support layer, fusion feature layer and output layer. The input end of the target detection algorithm (YOLOv5) uses the Mosaic algorithm to implement the design of data enhancement, the purpose is to enrich the data set and reduce the training time to a certain extent. In the support layer, the input image is cut into four parts using the focus structure. The data of each of the four parts is equivalent to that obtained under two times downsampling, and then the convolution operation is performed after the vertical channel splicing. In the fusion feature layer, the path aggregation network (PANet) is responsible for feature fusion. What it does is to add the flow path of information on the basis of the feature extractor (FPN), which shortens the information transmission path to a certain extent.

Before using the YOLOv5 model for pest identification and detection, the YOLOv5 model is trained first.

The ImageNet image dataset is used to train the YOLOv5 model. The dataset is divided into training set, validation set, and test set, which helps to evaluate the performance of the model and adjust the hyperparameters. The training set is used to update and optimize the model parameters, the validation set is used to monitor the performance of the model and select the best model configuration, and the test set is used to evaluate the performance of the final trained model on unseen data. In order to achieve accurate detection and identification of farmland pests, the multi-target detection model based on the YOLOv5 algorithm is used for model optimization based on the preprocessed safety behavior data.

In order to obtain better model performance, the stochastic gradient descent (SGD) algorithm is used to update the model parameters and gradually find the minimum value of the loss function, thereby improving the accuracy and effect of the model. The model optimization formula is expressed as follows:

;

Among them, y represents the true value of the label, represents the predicted value of the model, Represents the true value of the target position coordinates, Represents the model's predicted value for the target location; and Represent classification loss and positioning loss respectively, is the weight of the loss term.

The parameter update formula (using the SGD algorithm) is as follows: ;

in, represents the model parameters, represents the learning rate, It represents the gradient of the loss function with respect to the parameters. When training and optimizing the YOLOv5 model, it is necessary to comprehensively consider the data set, model configuration, training settings, training process, hyperparameter adjustment, model evaluation, model optimization and application. Through continuous trial and adjustment, better model performance and practical application effects can be obtained.

The image obtained after preprocessing in step 2 is used as the input of the optimized YOLOv5 model. For each frame of the input image, it is represented as x;

The output of the optimized YOLOv5 model includes the coordinate information of the target pest location and the category probability; for each input image x, the output is: ;

Among them, m represents the number of targets detected in the image. Represents the prediction result of the i-th target; each target prediction result Bounding box coordinates including the target location And the corresponding class probability , the target position coordinate formula is:

;

in, represents the coordinates of the upper left corner of the bounding box, , Represent the width and height of the bounding box respectively; the category probability formula is as follows:

;

Where c represents the number of target pest categories, Indicates the probability that the i-th target belongs to the j-th category. To identify the weight of pests, it is necessary to determine the type of each target based on the target category probability and the predefined pest type label, and take corresponding management measures. It should be noted that to achieve pest type identification, it is necessary to have certain basic knowledge of computer vision and deep learning, and understand the principles and applications of target detection algorithms such as YOLOv5. In practical applications, the generalization ability and real-time performance of the model should be considered, and optimization and adjustment should be carried out according to the actual situation.

Step 4: Optimize the image obtained in step 3 using the new enhanced quantum representation model.

Step 4 specifically includes:

Step 4.1, convert the digital image into a quantum image: the image size is , the grayscale range is , then the NEQR model is expressed as follows:

;

In the formula, is the image quantum state;

;

image It consists of two parts, namely and Respectively represent the grayscale information and position information of the stored image; position information It consists of vertical coordinates and horizontal coordinates, and the specific expression is:

;

Among them, the first n quantum bits store the vertical coordinate information of the image, and the last n quantum bits store the horizontal coordinate information of the image;

The NEQR model representation of a binary image is:

;

In the formula, Indicates the binary information of the stored image; Indicates the location information of the stored image;

Step 4.2, use quantum logic gates to transform the storage state: the initial state of the quantum system is:

;

Applying H-gate transformation, each qubit representing position information is transformed by H-gate transformation, which can generate The superposition system of states is:

;

According to the pixel information of the actual image and the mapping relationship between coordinates and grayscale information, the initial state sequence representing the grayscale information in the quantum image is converted into the corresponding M state, realizing the unique mapping process between the position information and the grayscale information in the quantum image. The specific expression is:

;

Where n represents the length of the quantum bit; It represents all the ground states in the quantum grayscale image system; Represents tensor product calculation in quantum computing; Represents binary grayscale information that can be used for encoding; Represents a grayscale image;

Step 4.3, perform quantum measurement on the image quantum state: position information measurement operator for:

;

In the formula, represents the direct product of q identity matrices;

The grayscale value from the quantum state is recovered using projection measurement. The measurement operator M of the grayscale information is:

;

in, and Both represent the location information of the stored image.

Step 5: extract the data features of the pest images, classify the pests with the same data features, and establish a farmland pest classification database.

Table 1 Degree of pest damage

Most insects in farmland can be divided into beneficial insects and pests according to the benefits and harms of insects to farmland. After processing the image with the NEQR model, insects can be screened and pests can be classified according to their shape, color and texture.

Beneficial insects: Beneficial insects in farmland mainly include predatory natural insects, parasitic natural enemy insects, insect pathogenic microorganisms, and pollinating insects.

Pests: Plants will inevitably be infected by pests during cultivation. If the pests and diseases are not prevented and controlled in time, it will cause great harm to agricultural production. Common pests that can be screened out after processing images based on the NEQR model include small cutworms, cabbage armyworms, wireworms, corn borers, armyworms, wheat stalk flies, sorghum aphids, and grass borers for research.

According to the research, the degree of damage to farmland by several common pests selected is mainly divided into five levels: light, medium light, medium, medium severe, and severe. The selected farmland crops are different, the pest types are different, and the threshold of the degree of damage to farmland is also different. For example, as shown in Table 1, the degree of damage of small cutworms in corn farmland is light, medium light, medium, medium severe, and the corresponding number of small cutworms is <2, 2-6, 6-11, 11-16, >16 (the number of pests per 100 corn plants).

It can be seen that the farmland with pests is absolute, the degree of pests is relative, and the incidence of crops affected by various pests is different. Prevention of pests and diseases is more effective at a medium or low level. For example, when the number of armyworms is 150 per square meter, the number of wheat stalk flies is 1-3 per 200 nets, the aphid rate of sorghum aphids is 30%-50% per 100 plants, the number of cutworms is 2-6 per 100 plants, and the number of grass borers is 30-70 per square meter, prevention is effective.

Through online search and field investigation, 10 kinds of farmland pest images, a total of 2386 samples, were sorted out and annotated. Data enhancement operations such as random flipping and random contrast adjustment of the original images were adopted to enhance the data diversity, thereby achieving the purpose of expanding the collected sample set, improving the robustness of the target detection model, and being more adaptable to real data. The farmland pest dataset was completed, and the dataset produced can be used in research fields related to farmland pest detection.

Step 6: Collect images of farmland with pests, use pixel grayscale association to detect image edges, determine the types and numbers of pests in each area of the farmland based on the farmland pest classification database obtained in step 5, and spray corresponding pesticides in a centralized manner.

In step 6, the image edge detection using pixel grayscale association specifically includes:

Step 6.1, convert the image into a quantum image:

Digital image after grayscale processing Indicates at point The gray value at , ;like Represents pixel The probability that the gray value is 1, is the probability that the gray value is 0, then the image The available quantum bits can be represented as:

;

Step 6.2, describe the grayscale transition of the neighboring pixels: , and The gray values of three adjacent pixels are , and , according to the principle of quantum state superposition, they constitute a neighborhood grayscale correlation system, expressed as:

;

In the image quantum correlation system, the state vector is composed of three binary bits. From left to right, the state vector with a jump of 01 is a positive state vector, and the state vector with a jump of 10 is a negative state vector.

According to the difference operation between the probability of positive and negative state vectors and the probability of their corresponding anti-state vectors, we have

;

in, is the superposition vector The probability amplitudes corresponding to the previous coefficients respectively;

Step 6.3, construct edge detection template using average formula:

The grayscale transition of the image is described by the sum of the detection components in the horizontal and vertical directions, that is, the edge grayscale: ;

in, Indicates the image transition, that is, the edge grayscale, represents the detection component in the horizontal direction, Represents the detection component in the vertical direction;

Step 6.4, edge detection threshold setting:

The triple mean of the edge grayscale is used as the edge detection threshold, that is:

;

in, represents the threshold for edge detection, Represents the edge grayscale mean;

Extract binary edge image function ; If the edge grayscale mean is greater than the edge detection threshold, the binary edge image function is 0, otherwise the binary edge image edge function is 1.

According to the obtained threshold data, the body shape, characteristics, color and other data features of the pest image are extracted to establish a farmland pest classification database, and the degree of damage caused by different types of pests to farmland is analyzed and classified. In addition, a pest area calibration is established to determine the type and number of pests in each area, and the corresponding pesticides are sprayed in a centralized manner to achieve the purpose of more accurately reducing farmland pests and increasing crop yields.

Compared with the traditional training model, this embodiment improves the average accuracy of the whole class by 8.1% and the recall rate by 12.8%. Since the quantum-improved deep learning target detection algorithm (YOLOv5) model enhances the model's ability to extract and transmit pest area information, and secondly, by optimizing the loss function, the model's attention to pest samples in complex environments is improved. Therefore, the quantum-improved deep learning target detection algorithm (YOLOv5) effectively extracts important pest features in complex farmland environments and achieves good detection results. The quantum-improved deep learning target detection algorithm (YOLOv5) strengthens the representation of pest features in the deep layer, improves the accuracy of multi-scale pest detection, and achieves more accurate detection with a small number of parameters. Pests can be accurately located, making the coverage of pest areas tighter and the positioning accuracy higher.

The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims (4)

1. A method for detecting farmland pests based on quantum deep learning, characterized in that it specifically includes the following steps: Step 1, collecting images of detected pests, and converting the collected images of detected pests into quantum images using a novel enhanced quantum representation model; Step 2, preprocessing the pest image obtained in step 1, and storing the image position information in the quantum system; Step 3: Build an improved deep convolutional neural network to classify and identify pest images. If they are pests, input the pest image into the image segmentation network to reconstruct the pest image. Step 4, optimizing the image obtained in step 3 using the new enhanced quantum representation model; Step 5, extracting data features of pest images, and classifying pests with the same data features to establish a farmland pest classification database; Step 6: collect images of farmland with pests, perform image edge detection using pixel grayscale association, determine the types and numbers of pests in each area of the farmland based on the farmland pest classification database obtained in step 5, and spray corresponding pesticides in a centralized manner; The step 3 specifically includes: using the preprocessed image obtained in step 2 as the input of the optimized YOLOv5 model, and representing the input image of each frame as x; The output of the optimized YOLOv5 model includes the coordinate information of the target pest location and the category probability; for each input image x, the output is y = [y ₁ ,y ₂ ,…,y _m ] Where m represents the number of targets detected in the image, _yi represents the prediction result of the ith target; each target prediction result _yi includes the bounding box coordinates of the target position ( _bi ) and the corresponding category probability ( _pi ), and the target position coordinate formula is: b _i =(xi _{, yi} , _wi , _hi ) Among them, ( _xi , _yi ) represents the coordinates of the upper left corner of the bounding box, _wi , _hi represent the width and height of the bounding box respectively; the category probability formula is as follows: p _i =(p _{i1} ,p _{i2} ,…,p _{ic} ) Where c represents the number of target pest categories, p _{ic} represents the probability that the i-th target belongs to the j-th category; The image edge detection using pixel grayscale association in step 6 specifically includes: Step 6.1, convert the image into a quantum image: The grayscale processed digital image f(x,y) represents the grayscale value at the point (x,y), f(x,y)∈[0,1],(x,y)∈Z ² ; if f(x,y) represents the probability that the grayscale value at the pixel point (x,y) is 1, and 1-f(x,y) is the probability that the grayscale value is 0, then the image f(x,y) can be represented by quantum bits as: Step 6.2, describe the transition of the grayscale of the neighboring pixels: the grayscale values of the three neighboring pixels (x,y-1), (x,y) and (x,y+1) in the image are fx _,y-1 , _fx,y and _fx,y+1 respectively. According to the principle of quantum state superposition, they constitute a neighborhood grayscale correlation system, which can be expressed as: In the image quantum correlation system, the state vector is composed of three binary bits. From left to right, the state vector with a jump of 01 is a positive state vector, and the state vector with a jump of 10 is a negative state vector. According to the difference operation between the probability of positive and negative state vectors and the probability of their corresponding anti-state vectors, we have (|ω ₀₁₁ (x,y)| ² -|ω ₁₀₁ (x,y)| ² )=(f _x,y -f _x,y-1 )f _x,y+1 (|ω ₀₀₁ (x,y)| ² -|ω ₀₁₀ (x,y)| ² )=(1-f _x,y-1 )(f _x,y+1 -f _x,y ) (|ω ₁₀₀ (x,y)| ² -|ω ₀₁₀ (x,y)| ² )=(f _x,y-1 -f _x,y )(1-f _x,y+1 ) (|ω ₁₁₀ (x,y)| ² -|ω ₁₀₁ (x,y)| ² )＝f _x,y-1 (f _x,y -f _x,y+1 ) Among them, ω _i (x, y) is the probability amplitude corresponding to the coefficients before the superposition state vector |i>; Step 6.3, construct edge detection template using average formula: The grayscale transition of the image is described by the sum of the detection components in the horizontal and vertical directions, that is, the edge grayscale H(x,y)＝|H ₀ (x,y)|+|H _p (x,y)| Among them, H(x,y) represents the image transition, that is, the edge grayscale, _H0 represents the detection component in the horizontal direction, and _Hp represents the detection component in the vertical direction; Step 6.4, edge detection threshold setting: The triple mean of the edge grayscale is used as the edge detection threshold, that is: thr＝3×mean(H(x,y)) In the above formula, thr represents the threshold of edge detection, and mean(H(x,y)) represents the mean grayscale value of the edge. Extract the binary edge image function f ₀ ; if the edge grayscale mean is greater than the edge detection threshold, the binary edge image function is 0, otherwise the binary edge image function is 1. 2. According to the method for detecting farmland pests based on quantum deep learning in claim 1, it is characterized in that the expression of the novel enhanced quantum representation model in step 1 is: Among them, |Q> represents a grayscale image; |M> represents binary grayscale information that can be used for encoding; n represents the length of the quantum bit; Represents a tensor product calculation in quantum computing. 3. According to claim 1, a method for detecting farmland pests based on quantum deep learning is characterized in that the deep convolutional neural network in step 3 is an optimized YOLOv5 model, and the optimized YOLOv5 model includes an input end, a support layer, a fusion feature layer and an output layer, and the input end uses a mosaic algorithm to achieve data enhancement; the support layer uses a focusing structure to cut the input image, cuts the image into four parts, and performs a convolution operation after splicing through longitudinal channels; the path aggregation network in the fusion feature layer is responsible for feature fusion, and adds an information flow path based on the feature extractor; during the model training process, the stochastic gradient descent algorithm is used to update the model parameters. 4. The method for detecting farmland pests based on quantum deep learning according to claim 1, characterized in that step 4 specifically comprises: Step 4.1, convert the digital image into a quantum image: the image size is ²ⁿ × ²ⁿ , the grayscale range is [0, ^2q-1 ], then the NEQR model is expressed as follows: Where |I> is the image quantum state; The image |I> consists of two parts, namely and |YX> represent the grayscale information and position information of the stored image respectively; the position information |YX> consists of vertical coordinates and horizontal coordinates, as shown in the following formula: |YX>＝|y _n-1 ,y _n-2 ,…,y ₀ >|x _n-1 ,x _n-2 ,…,x ₀ >,x _i ,y _i ∈{0,1} Among them, the first n quantum bits store the vertical coordinate information of the image, and the last n quantum bits store the horizontal coordinate information of the image; The NEQR model representation of a binary image is: G _YX ∈{0,1},Y,X＝0,1,…,2 ⁿ -1 In the formula, |G _YX > stores the binary information of the image; |YX> stores the position information of the image; Step 4.2, use quantum logic gates to transform the storage state: the initial state of the quantum system is By applying the H-gate transformation, each qubit representing the position information is transformed by H, which can generate a superposition state system with ²ⁿ states, that is, According to the pixel information of the actual image, based on the mapping relationship between coordinates and grayscale information, the initial state sequence representing the grayscale information in the quantum image is converted into the corresponding M state, realizing the unique mapping process between the position information and the grayscale information in the quantum image, as shown in the following formula Where n represents the length of the quantum bit; |j>＝|0>,|1>,…|2 ²ⁿ -1> represents all the ground states in the quantum grayscale image system; represents the tensor product in quantum computing; |M> represents binary grayscale information that can be used for encoding; |F> represents a grayscale image; Step 4.3, perform quantum measurement on the image quantum state: the position information measurement operator Γ is In the formula, represents the direct product of q identity matrices; The grayscale value from the quantum state is recovered using projection measurement. The measurement operator M of the grayscale information is: Among them, |YX> and <YX| both represent the location information of the stored image.