增强细节信息特征提取的鱼类个体识别算法OACSTPCD

In practical application scenarios for fish individual recognition,due to high underwater environmental noise,tilted fish angles,and insignificant intra class feature differences,the feature extraction ability of convolutional neural networks is low,which affects the recognition accuracy.On this basis,The YOLOv5 network is improved and a loss function is established to optimize the detection results of fish individuals.Based on MobileNet-v1,the main network is optimized,deep convolution layers are improved,ReLU activation functions are updated,and Leaky ReLU is used to retain negative feature information for feature acquisition.A feature weighting layer is added before the fully connected layer at the end of the network structure to enhance the fine details in the images and extract features by removing the pooling layer commonly used in convolutional neural networks.The experimental results show that the proposed model achieves an average accuracy of 92.46%and a maximum accuracy of 95.69%on the DLOUFish dataset.