Automated Work Method for Oil Pipeline Submarine Trencher Based on Machine Vision

In the automation workflow of subsea trenching machines in oil pipeline installation, the detection of underwater targets forms the basis for intelligent operation. Underwater imaging conditions are complex, with low resolution and limited information for small targets, making it difficult to extract effective feature information. This leads to low detection and recognition rates for small underwater targets, as well as high false alarm rates. To address this issue, this paper proposes a machine vision-based multi-scale underwater small target detection method. This method efficiently promotes the automation of subsea trenching machines in oil pipeline installation. The method utilizes the DarkNet53 backbone network model for feature extraction to obtain high-level semantic information. It incorporates a multi-rate aurous convolution module to expand the network's receptive field. By adjusting the dilation rate, feature information is captured over a larger pixel range, and residual structures are added to preserve detailed information for small target localization. To restore the resolution of small targets, a deconvolution module is used to reconstruct image details and learn feature details from different resolution feature maps. Building upon this, a feature pyramid structure is introduced into the deconvolution layer to incorporate richer multi-scale contextual information, enabling cross-scale learning of features at multiple levels to enhance small target localization and classification. Each layer's output is then integrated and filtered to obtain the final prediction results. Experimental results demonstrate that this method effectively enhances the detection capability of small underwater targets while ensuring real-time detection.