A Gender Recognition Method for Chinese Mitten Crab in Low Light Environment Based on Machine Vision

The acquisition of gender information of Chinese mitten crabs is crucial for survival rate assessment and biomass estimation. Machine vision is commonly used for the automatic gender recognition of Chinese mitten crabs, but the low light conditions caused by deeper containers reduce the accuracy of existing algorithms. To improve the accuracy of gender recognition of Chinese mitten crabs in low light environments, this study incorporates the GAM into the backbone of YOLOv7, improving the capability to distinguish gender-specific features. A lightweight upsampling operator, CARAFE, replaces the original upsampling module, expanding the model's receptive field and better aggregating global information. Using Focal-Clo U to mitigate the impact of low-quality prediction boxes, enabling the model to better optimize high-quality ones. By constructing an image dataset of Chinese mitten crabs in low-light environments to validate the model's performance, experimental results demonstrate that the average precision of the model established in this study is 0.870, 10.6% higher than the most update algorithm GMNet-YOLOv4, effectively performing the classification task of Chinese mitten crabs in low-light environments.