关键词: 图像处理, 卷积神经网络, 主成分分析, 蚕茧分选

Abstract: In view of the inaccuracies and low efficiency of traditional manual classification in cocoon sorting, a cocoon recognition method based on an improved convolutional neural network and image processing was proposed. The methods model has 5 convolutional layers, 5 pooling layers, and 4 full connection layers. The method uses principal component analysis (PCA) and color space conversion to enhance the characteristics of the cocoons images. The experimental results show that the identification accuracy of the model is 96%, with the identification accuracy of Chaiyin cocoons increasing from 86.21% to 93.55%, and the identification accuracy of colored cocoons increasing from 96% to 100%. The results show that the image processing method can significantly improve the identification effect. Furthermore, the model meets the cocoon classification needs and provides an effective reference for actual production.

Key words: image processing, convolution neural network, principal component analysis, cocoon sorting