CN116563889A - Device and method for estimating weight of laying hen based on machine vision - Google Patents

Abstract

The invention discloses a machine vision-based laying hen weight estimation device and a machine vision-based laying hen weight estimation method, which belong to the technical field of machine vision and comprise a monitoring platform, an image acquisition module, an image enhancement module, a target extraction module, a weight estimation module, a parameter updating module, an alarm module and a block storage module; the invention can automatically model and search parameters, effectively improve the accuracy of weight estimation, reduce the operation difficulty, facilitate the use of breeding personnel, improve the use experience, accurately acquire the image data of the laying hens through cascade analysis, improve the target detection efficiency, reduce the workload of subsequent data processing and reduce the waiting time.

Description

A machine vision-based equipment and method for estimating the weight of laying hens

technical field

The invention relates to the technical field of machine vision, in particular to a device and method for estimating the body weight of laying hens based on machine vision.

Background technique

The aquaculture industry is an important branch of agriculture, mainly including poultry breeding, animal husbandry and aquaculture. Poultry farming mainly produces poultry meat and eggs; in recent years, intelligent farming has become an important research field in livestock and poultry farming, and in poultry farming, body weight has a greater impact on chicken meat and egg production performance, reflecting One of the important indicators of the feeding status of the chicken house. At the same time, real-time monitoring of poultry weight data can maximize the production and economic benefits of the farm, and has great reference value for poultry breeding, virus prevention and environmental control. With the continuous development of machine vision technology, the image analysis method based on machine vision has also become one of the important means of monitoring the weight of laying hens.

The existing equipment and methods for weight estimation of laying hens require manual modeling and parameter search. The accuracy of weight estimation is poor and the operation is difficult. In addition, the existing equipment and methods for weight estimation of laying hens have low target detection efficiency and subsequent data The processing workload is relatively large. For this reason, we propose a machine vision-based equipment and method for estimating the weight of laying hens.

Contents of the invention

The object of the present invention is to propose a machine vision-based laying hen body weight estimation device and method in order to solve the defects in the prior art.

In order to achieve the above object, the present invention adopts the following technical solutions:

A machine vision-based weight estimation device for laying hens, including a monitoring platform, an image acquisition module, an image enhancement module, a target extraction module, a weight estimation module, a parameter update module, an alarm module, and a block storage module;

The monitoring platform is used to verify the identity of the breeders and receive feedback data from each module for the breeders to check;

The image collection module is used to collect the image information of the breeding area;

The image enhancement module is used to optimize the collected image information of the breeding area;

The target extraction module is used to extract the image of laying hens in the image information of the breeding area;

The body weight estimation module is used for estimating the body weight of each laying hen according to the collected laying hen images;

The parameter update module is used to collect the operation information of the weight estimation module and update its parameters;

The alarm module is used to feed back abnormal laying hen information to the breeder, and at the same time mark the laying hen;

The block storage module is used to store the detection information of laying hens on the chain.

As a further solution of the present invention, the specific steps of optimizing the image information of the breeding area of the image enhancement module are as follows:

Step 1: Extract the collected image information of the breeding area frame by frame to obtain multiple sets of image data, and then perform block processing according to the display ratio of each image data, and then perform Fourier transform on each group of image data after block Analyze and extract the high-frequency components in the data, and smooth them through Gaussian filtering;

Step 2: Use a window of specified pixels to move in each group of image information, calculate the gray level co-occurrence matrix under the window at each time of movement, and calculate the texture features in the relevant image information from the gray level co-occurrence matrix.

As a further solution of the present invention, the specific calculation formula of the texture feature described in step 2 is as follows:

In the formula, P(i, j) represents the value of the image pixel; L represents the number of gray levels;

Among them, the formula (1) is used to calculate the texture feature energy, the energy moment of the coarse texture is larger, and the energy moment of the fine texture is smaller; the formula (2) is used to calculate the entropy value of the texture feature, if the image does not have any texture, the entropy is close to Zero, if the image is full of fine textures, the entropy value of the image is the largest, if there are fewer textures distributed in the image, the entropy value of the image is smaller; the formula (3) is used to calculate the texture feature contrast, the greater the contrast , the visual clarity of the image is better; the formula (4) is used to calculate the texture feature correlation, and its correlation is used to measure the linear dependence of the neighborhood gray level.

As a further solution of the present invention, the specific steps of the layer image extraction of the target extraction module are as follows:

Step ①: Scale normalize the optimized image information through the image pyramid, and extract the features of each group of image information, and then perform feature fusion through the bidirectional feature pyramid to obtain the target detection frame;

Step ②: Expand and crop each image information according to the target detection frame to obtain the target image, and then obtain the correlation and contrast of the texture features obtained by sliding the window, and store the obtained feature values in the corresponding pixel position in the form of an array;

Step ③: When the correlation and contrast meet the preset conditions, judge the current pixel area as the target laying hen, and mark it as 1; The image is subjected to background separation to extract the target layer image;

Step ④: Calculate the shape factor of the laying hen image, and select the laying hen image whose shape factor tends to 0, and judge that there are laying hen adhesion areas in the laying hen image, and estimate the number of laying hens in each adhesion area, and then perform a convex hull on the adhesion area Area filling, and then obtain the convex defect in the adhesion area according to the area of the convex hull and the area of the adhesion area to obtain the information of the concave point in the convex defect;

Step ⑤: For the adhesion of two chickens, directly draw a straight line through two concave points to divide the cohesive chicken body; for the adhesion of multiple individuals, randomly connect the detected concave points, and each concave point can only be connected once. Calculate the number of connected regions and the area of each connected region after each connection. When the area of each connected region is smaller than the maximum area, it is judged that the matching is complete, and the cohesive layer is divided according to the matching result.

As a further solution of the present invention, the specific steps of the body weight estimation module of the body weight estimation module are as follows:

Step Ⅰ: The weight estimation module receives the images of each group of laying hens, and extracts the edges of the binary images of each group of laying hens through the Roberts algorithm, and then uses the least square method to perform ellipse fitting on the edge contours, and at the same time, the fitted circle area The pixel value is filled, and then the circle area is subtracted from the binary image area, and a small area of the binary image outside the ellipse area is extracted;

Step Ⅱ: Extract the edge information of the extracted small area, and calculate the minimum distance from each edge point to the circle boundary, then extract the maximum value of these distances and the coordinates of the corresponding point, and then perform a mask centered on this point Eliminate, and repeat the above steps until the maximum distance is less than the preset target value and then stop;

Step Ⅲ: The weight estimation module extracts multiple sets of laying hen weight data from the block storage module, and then integrates them into a sample data set, and calculates the standard deviation of the sample data set to eliminate abnormal data in the sample data set, and then Standardize and normalize the remaining data;

Step Ⅳ: Divide the normalized data into training set, test set and verification set according to a certain ratio, then assign a set of parameter setting vectors of convolutional neural network, and then determine the number of neurons in each neural network layer and Corresponding to the activation function, the training set is input into the convolutional neural network for iterative training to obtain the estimated model;

Step Ⅴ: Verify the model accuracy of the predicted model through the verification set, and then test the performance of the model with the input value of the test set, and calculate its evaluation index. If the evaluation index is greater than the specified threshold, then use the parameter adjustment module to adjust The estimated model parameters are updated;

Step Ⅵ: After the update is completed, enter the image of the layered hen after the adhesion and segmentation into the estimation model, and output the estimated weight value, and record the estimated weight value and average weight value of multiple times at the same time, and record the number of the laying hens whose weight reaches the standard .

A method for estimating the body weight of laying hens based on machine vision, the estimating method is specifically as follows:

(1) Collect and optimize the image information of the breeding area;

(2) Calculate and record the image texture features of the breeding area;

(3) Extract the laying hen image in the breeding area image and carry out the adhesion segmentation;

(4) Construct an estimation model and optimize the model parameters;

(5) Estimate and record the body weight of each laying hen by the estimation model;

(6) Report the weight of abnormal layers and store the estimated information on the chain.

As a further solution of the present invention, the specific steps of parameter optimization of the estimated model described in step (4) are as follows:

Step I: The parameter update module initializes the network connection weights within the specified interval of the estimated model, then submits training samples from the set of input and output pairs during training, and calculates the output of the estimated model, and then compares the expected The network output and the actual network output, and calculate the local error of all neurons;

Step II: When the local error exceeds the preset threshold, train and update the weight of the estimated model according to the learning rule equation, and list all possible data results according to the preset learning rate and step size;

Step III: For each set of data, select any subset as the test set, and the remaining subsets as the training set, test the test set after training the test model, and count the root mean square error of the test results;

Step IV: Replace the test set with another subset, then take the remaining subset as the training set, and count the root mean square error again until all the data are predicted once, by selecting the combination corresponding to the minimum root mean square error The parameters are used as the optimal parameters in the data interval and replace the original parameters of the prediction model.

Compared with the prior art, the beneficial effects of the present invention are:

1. The system receives the images of each group of laying hens through the weight estimation module, and performs ellipse fitting on the images of each group of laying hens. The weight estimation module extracts multiple sets of laying hen weight data from the block storage module for preprocessing and Data division, the training set is input into the convolutional neural network for iterative training to obtain the estimated model, the model accuracy of the estimated model is verified through the verification set, and then the test set is input to the estimated model to test the performance of the model and calculate its Evaluation index, if the evaluation index is greater than the specified threshold, the parameter adjustment module updates the parameters of the estimated model through the cross-validation method. After the update is completed, the image of the layer after the adhesion and segmentation is entered into the estimated model, and the estimated weight value is output. Simultaneously record multiple estimated weight values and average weight values, and record the number of laying hens whose weight reaches the standard. It can model and find references by itself, effectively improving the accuracy of weight estimation, while reducing the difficulty of operation, and facilitating the use of farmers. Improve user experience.

2. The present invention performs scale normalization processing on the optimized image information through the image pyramid, and extracts the features of each group of image information, and then performs feature fusion through the two-way feature pyramid to obtain the target detection frame, and performs the detection of each image according to the target detection frame. The information is enlarged and clipped to obtain the target image, and then the correlation and contrast of the texture features obtained by sliding the window are obtained, and the background of the target image is separated to extract the target layer image, and the number of layers in each adhesion area is estimated, and then the adhesion area is analyzed. Fill the convex hull area, and then obtain the convex defect in the adhesion area according to the area of the convex hull and the area of the adhesion area to obtain the information of the concave points in the convex defect. For the adhesion of two chickens, directly draw a straight line passing through the two concave points to divide the cohesive chicken body; For the adhesion of multiple individuals, the detected concave points are randomly connected, and each concave point can only be connected once. After each connection, the number of connected regions and the area of each connected region are calculated. When each connected region When the areas of the layers are all smaller than the maximum area, it is judged that the matching is complete, and the cohesive laying hens are segmented according to the matching results. Through the cascade analysis, the image data of the existing laying hens can be accurately obtained, the target detection efficiency is improved, and the subsequent data processing work is reduced. volume, reducing waiting time.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the description, and are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention.

Fig. 1 is a system block diagram of a machine vision-based laying hen body weight estimation device proposed by the present invention;

Fig. 2 is a flow chart of a method for estimating the body weight of laying hens based on machine vision proposed by the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention.

Example 1

With reference to Fig. 1, a kind of laying hen weight estimation equipment based on machine vision, comprises monitoring platform, image acquisition module, image enhancement module, target extraction module, weight estimation module, parameter update module, alarm module and block storage module;

The monitoring platform is used to verify the identity of the farming personnel and receive feedback data from each module for viewing by the farming personnel; the image acquisition module is used to collect the image information of the farming area.

The image enhancement module is used to optimize the collected image information of the breeding area.

Specifically, the collected image information of the breeding area is extracted frame by frame to obtain multiple sets of picture data, and then block processing is performed according to the display ratio of each picture data, and then each set of picture data after block is processed by Fourier transform Analyze and extract the high-frequency components in the data, and smooth them through Gaussian filtering, use the window of specified pixels to move in each group of image information, and calculate the gray level co-occurrence matrix under the window at this time every time you move, and calculate from the gray level co-occurrence matrix Texture features in associated image information.

It needs to be further explained that the specific calculation formula of texture features is as follows:

In the formula, P(i, j) represents the value of the image pixel; L represents the number of gray levels;

Among them, the formula (1) is used to calculate the texture feature energy, the energy moment of the coarse texture is larger, and the energy moment of the fine texture is smaller; the formula (2) is used to calculate the entropy value of the texture feature, if the image does not have any texture, the entropy is close to Zero, if the image is full of fine textures, the entropy value of the image is the largest, if there are fewer textures distributed in the image, the entropy value of the image is smaller; the formula (3) is used to calculate the texture feature contrast, the greater the contrast , the visual clarity of the image is better; the formula (4) is used to calculate the texture feature correlation, and its correlation is used to measure the linear dependence of the neighborhood gray level.

The target extraction module is used to extract the image of laying hens in the image information of the breeding area.

Specifically, the optimized image information is scale-normalized through the image pyramid, and the features of each group of image information are extracted, and then feature fusion is performed through the bidirectional feature pyramid to obtain the target detection frame, and each image information is analyzed according to the target detection frame. Expand and crop to obtain the target image, and then obtain the correlation and contrast of the texture features obtained by sliding the window, and store the obtained feature values in the corresponding pixel position in the form of an array. When the correlation and contrast meet the preset conditions, Then it is judged that the current pixel area is the target layer, and it is marked as 1. If it is not satisfied, the current pixel area is judged to be the background area, and the background of the target image is separated according to the judgment result to extract the target layer image, and the layer is calculated The shape factor of the image, and select the layer image whose shape factor tends to 0, and judge that there is a layer cohesion area in the layer image, and estimate the number of layers in each cohesion area, and then fill the cohesion area with a convex hull area, and then according to the convex hull In order to obtain the concave point information in the convex defect, for the adhesion of two chickens, directly draw a straight line passing through the two concave points to divide the cohesive chicken body; for the adhesion of multiple individuals, the The detected concave points are randomly connected, and each concave point can only be connected once. After each connection, the number of connected regions and the area of each connected region are calculated. When the area of each connected region is smaller than the maximum area, Then it is judged that the matching is completed, and the sticky laying hens are segmented according to the matching result.

The body weight estimation module is used for estimating the body weight of each layer hen according to the collected layer images.

Specifically, the weight estimation module receives each group of laying hen images, and extracts the binary image edges of each group of laying hen images through the Roberts algorithm, and then uses the least square method to perform ellipse fitting on the edge contours, and at the same time, the fitted circle area The pixel value is filled, and then the circle area is subtracted from the binary image area, and the small area of the binary image outside the ellipse area is extracted, the edge information of the extracted small area is extracted, and each edge point is calculated to the circle boundary The minimum distance, and then extract the maximum value of these distances and the coordinates of the corresponding point, and then remove the mask centered on this point, and then repeat the above steps until the maximum distance is less than the preset target value and stop. The weight estimation module starts from The block storage module extracts multiple sets of laying hen weight data, and then integrates them into a sample data set, and calculates the standard deviation of the sample data set to eliminate abnormal data in the sample data set, and then standardizes and normalizes the remaining data. Normalization processing divides the normalized data into training set, test set and verification set according to a certain ratio, and then assigns a set of parameter setting vectors of convolutional neural network, and then determines the number of neurons in each neural network layer And the corresponding activation function, input the training set into the convolutional neural network for iterative training to obtain the estimated model, verify the model accuracy of the estimated model through the verification set, and then test the performance of the model by inputting the value of the test set into the estimated model, And calculate its evaluation index. If the evaluation index is greater than the specified threshold, the parameters of the estimated model are updated through the parameter adjustment module. After the update is completed, the image of the layer after the adhesion and segmentation is entered into the estimated model, and the estimated weight value is output , record multiple estimated weight values and average weight values at the same time, and record the number of laying hens whose weight reaches the standard.

The parameter update module is used to collect the operation information of the weight estimation module and update its parameters; the alarm module is used to feed back the abnormal layer information to the breeders, and at the same time mark the layer; the block storage module is Chicken detection information is stored on the chain.

Example 2

Referring to Figure 2, a method for estimating the body weight of laying hens based on machine vision, the estimating method is as follows:

Collect and optimize the image information of the breeding area.

Calculate and record the image texture features of the farming area.

The laying hen images in the breeding area images are extracted and segmented by adhesion.

Build a prediction model and optimize the model parameters.

Specifically, the parameter update module initializes the network connection weights within the specified range of the estimated model, then submits training samples from the set of input and output pairs during training, calculates the output of the estimated model, and compares the expected network output and the actual network output, and calculate the local errors of all neurons. When the local errors exceed the preset threshold, the weights of the estimated model are trained and updated according to the learning rule equation, and the preset learning rate and The step size lists all possible data results. For each set of data, select any subset as the test set, and the remaining subsets as the training set. After training the test model, test the test set, and count the root mean square of the test results Error, replace the test set with another subset, then take the remaining subset as the training set, and count the root mean square error again until all the data are predicted once, by selecting the combination parameter corresponding to the minimum root mean square error as The optimal parameters in the data interval and replace the original parameters of the prediction model.

Estimate and record the body weight of each laying hen through the estimation model.

Report abnormal laying hen weight and store the estimated information on the chain.

Claims (7)

1. A machine vision-based laying hen weight estimation device, characterized in that it includes a monitoring platform, an image acquisition module, an image enhancement module, a target extraction module, a weight estimation module, a parameter update module, an alarm module and a block storage module; The monitoring platform is used to verify the identity of the breeders and receive feedback data from each module for the breeders to check; The image collection module is used to collect the image information of the breeding area; The image enhancement module is used to optimize the collected image information of the breeding area; The target extraction module is used to extract the image of laying hens in the image information of the breeding area; The body weight estimation module is used for estimating the body weight of each laying hen according to the collected laying hen images; The parameter update module is used to collect the operation information of the weight estimation module and update its parameters; The alarm module is used to feed back abnormal laying hen information to the breeder, and at the same time mark the laying hen; The block storage module is used to store the detection information of laying hens on the chain. 2. a kind of laying hen body weight estimation equipment based on machine vision according to claim 1, is characterized in that, described image enhancement module breeding area image information optimization specific steps are as follows: Step 1: Extract the collected image information of the breeding area frame by frame to obtain multiple sets of image data, and then perform block processing according to the display ratio of each image data, and then perform Fourier transform on each group of image data after block Analyze and extract the high-frequency components in the data, and smooth them through Gaussian filtering; Step 2: Use a window of specified pixels to move in each group of image information, calculate the gray level co-occurrence matrix under the window at each time of movement, and calculate the texture features in the relevant image information from the gray level co-occurrence matrix. 3. A kind of laying hen body weight estimation equipment based on machine vision according to claim 2, is characterized in that, the concrete calculation formula of texture feature described in step 2 is as follows: In the formula, P(i, j) represents the value of the image pixel; L represents the number of gray levels; Among them, the formula (1) is used to calculate the texture feature energy, the energy moment of the coarse texture is larger, and the energy moment of the fine texture is smaller; the formula (2) is used to calculate the entropy value of the texture feature, if the image does not have any texture, the entropy is close to Zero, if the image is full of fine textures, the entropy value of the image is the largest, if there are fewer textures distributed in the image, the entropy value of the image is smaller; the formula (3) is used to calculate the texture feature contrast, the greater the contrast , the visual clarity of the image is better; the formula (4) is used to calculate the texture feature correlation, and its correlation is used to measure the linear dependence of the neighborhood gray level. 4. a kind of laying hen body weight estimation equipment based on machine vision according to claim 2, is characterized in that, described target extraction module laying hen image extraction specific steps are as follows: Step ①: Scale normalize the optimized image information through the image pyramid, and extract the features of each group of image information, and then perform feature fusion through the bidirectional feature pyramid to obtain the target detection frame; Step ②: Expand and crop each image information according to the target detection frame to obtain the target image, and then obtain the correlation and contrast of the texture features obtained by sliding the window, and store the obtained feature values in the corresponding pixel position in the form of an array; Step ③: When the correlation and contrast meet the preset conditions, judge the current pixel area as the target laying hen, and mark it as 1; The image is subjected to background separation to extract the target layer image; Step ④: Calculate the shape factor of the laying hen image, and select the laying hen image whose shape factor tends to 0, and judge that there are laying hen adhesion areas in the laying hen image, and estimate the number of laying hens in each adhesion area, and then perform a convex hull on the adhesion area Area filling, and then obtain the convex defect in the adhesion area according to the area of the convex hull and the area of the adhesion area to obtain the information of the concave point in the convex defect; Step ⑤: For the adhesion of two chickens, directly draw a straight line through two concave points to divide the cohesive chicken body; for the adhesion of multiple individuals, randomly connect the detected concave points, and each concave point can only be connected once. Calculate the number of connected regions and the area of each connected region after each connection. When the area of each connected region is smaller than the maximum area, it is judged that the matching is complete, and the cohesive layer is divided according to the matching result. 5. A kind of laying hen body weight estimation equipment based on machine vision according to claim 4, is characterized in that, described body weight estimation module laying hen body weight estimation specific steps are as follows: Step Ⅰ: The weight estimation module receives the images of each group of laying hens, and extracts the edges of the binary images of each group of laying hens through the Roberts algorithm, and then uses the least square method to perform ellipse fitting on the edge contours, and at the same time, the fitted circle area The pixel value is filled, and then the circle area is subtracted from the binary image area, and a small area of the binary image outside the ellipse area is extracted; Step Ⅱ: Extract the edge information of the extracted small area, and calculate the minimum distance from each edge point to the circle boundary, then extract the maximum value of these distances and the coordinates of the corresponding point, and then perform a mask centered on this point Eliminate, and repeat the above steps until the maximum distance is less than the preset target value and then stop; Step Ⅲ: The weight estimation module extracts multiple sets of laying hen weight data from the block storage module, and then integrates them into a sample data set, and calculates the standard deviation of the sample data set to eliminate abnormal data in the sample data set, and then Standardize and normalize the remaining data; Step Ⅳ: Divide the normalized data into training set, test set and verification set according to a certain ratio, then assign a set of parameter setting vectors of convolutional neural network, and then determine the number of neurons in each neural network layer and Corresponding to the activation function, the training set is input into the convolutional neural network for iterative training to obtain the estimated model; Step Ⅴ: Verify the model accuracy of the predicted model through the verification set, and then test the performance of the model with the input value of the test set, and calculate its evaluation index. If the evaluation index is greater than the specified threshold, then use the parameter adjustment module to adjust The estimated model parameters are updated; Step Ⅵ: After the update is completed, enter the image of the layered hen after the adhesion and segmentation into the estimation model, and output the estimated weight value, and record the estimated weight value and average weight value of multiple times at the same time, and record the number of the laying hens whose weight reaches the standard . 6. A method for estimating the body weight of laying hens based on machine vision, characterized in that, the method for estimating is as follows: (1) Collect and optimize the image information of the breeding area; (2) Calculate and record the image texture features of the breeding area; (3) Extract the laying hen image in the breeding area image and carry out the adhesion segmentation; (4) Construct an estimation model and optimize the model parameters; (5) Estimate and record the body weight of each laying hen by the estimation model; (6) Report the weight of abnormal layers and store the estimated information on the chain. 7. a kind of laying hen body weight estimation method based on machine vision according to claim 6, is characterized in that, the detailed steps of estimation model parameter optimization described in step (4) are as follows: Step I: The parameter update module initializes the network connection weights within the specified interval of the estimated model, then submits training samples from the set of input and output pairs during training, and calculates the output of the estimated model, and then compares the expected The network output and the actual network output, and calculate the local error of all neurons; Step II: When the local error exceeds the preset threshold, train and update the weight of the estimated model according to the learning rule equation, and list all possible data results according to the preset learning rate and step size; Step III: For each set of data, select any subset as the test set, and the remaining subsets as the training set, test the test set after training the test model, and count the root mean square error of the test results; Step IV: Replace the test set with another subset, then take the remaining subset as the training set, and count the root mean square error again until all the data are predicted once, by selecting the combination corresponding to the minimum root mean square error The parameters are used as the optimal parameters in the data interval and replace the original parameters of the prediction model.