A method of dense occlusion juvenile abalone segmentation based on improved coordinate attention mechanism

Abstract The detection and classification of juvenile abalone is a key link in the abalone breeding industry, and juvenile abalone are small, densely distributed and prone to complex occlusion, which makes the efficiency of manual measurement of biomass low and error large. Based on the framework of deep learning network, this paper proposes a method of dense juvenile abalone segmentation with improved coordinate attention mechanism, and realizes the high-efficiency non-contact segmentation detection of dense juvenile abalone in complex situations. The improved shape feature attention mechanism designed by the algorithm model makes the network pay more attention to the details of juvenile abalone by emphasizing or suppressing the importance of features. At the same time, the algorithm model also constructs a Transformer module embedded with a double-layer Gaussian prior encoder, which improves the accuracy of juvenile abalone feature prediction by replacing the feedforward network with a Gaussian prior. Finally, the segmentation data set of juvenile abalone (JASD) is established and the effectiveness of the proposed algorithm model is verified by experiments.