天津职业技术师范大学学报

针对人眼识别犬类准确率不高且效率低的问题，对基于AlexNet网络模型的图像识别分类算法进行了改进。在传统AlexNet网络结构的基础上，部分筛减了卷积层，并在2层全连接层后增加了Dropout层及L2正则化以进一步减少过拟合，添加了通道维度的注意力机制。实验结果表明，本研究算法的准确率较传统算法提高了2%～5%，得到了一种能有效进行犬类图像识别分类的解决方案。

Aiming at the problem of low accuracy and low efficiency in canine recognition by human eyes,this study tries to improve the image recognition and classification algorithm based on the AlexNet network model. On the basis of the traditional AlexNet network structure,the convolutional layer was partially reduced,the Dropout layer was added behind the two full connection layers,and the L2 regularization was added to further reduce the overfitting. The attention mechanism of channel dimension was added. The experimental results show that the accuracy of the proposed algorithm is about 2% to 5% higher than that of the traditional one. A solution that can effectively recognize and classify canine images is obtained.