一种融合CBAM注意力机制和金字塔卷积的经济鱼类识别方法

A Method for Economic Fish Identification That Integrates CBAM Attention Mechanism and Pyramid Convolution

海洋经济鱼类图像资源相对较少,导致神经网络训练效果较差,难以对海洋经济鱼类形成有效监管。通过网络抓取并通过数据增广增加图像数量,以ResNet50为基础网络框架,引入卷积块模块注意力机制(CBAM)并且将基础网络中普通卷积替换成金字塔卷积(PyConv)。利用该模型对比目鱼、马鲛鱼、鲻鱼、海鲈鱼、黑鲷鱼和金枪鱼6种常见的海洋经济鱼类进行分类识别,实验结果表明,对于比目鱼、鲻鱼、海鲈鱼和金枪鱼的分类精确率达到了100%,对于马鲛鱼的分类精确率为98.4%,黑鲷鱼的分类精确率为98.3%。实验证明改进后的模型具有较高的识别精度。

The image resources of marine economic fish are relatively small,resulting in poor neural network training effect,and it is difficult to form effective supervision of marine economic fish.Based on this problem,the number of images is increased through network capture and data expansion.Taking ResNet50 as the basic network framework,the convolution block module attention mechanism(CBAM)is introduced and the common convolution in the basic network is replaced by pyramid convolution(PyConv).The model is used to classify and identify six common marine economic fish:flounder,mackerel,mullet,sea perch,blackhead seabream,and tuna.The experimental results showed that the classification accuracy of flounder,mullet,sea perch and tuna reached 100%,the classification accuracy of mackerel was 98.4%,and the classification accuracy of blackhead seabream was 98.3%.Experiments show that the improved model has high recognition accuracy.