| Cocoons are high-grade textile raw materials.In order to ensure the quality of textiles,cocoons are classified and identified before silk reeling.In recent years,as silk reeling enterprises have higher and higher requirements for the classification and recognition accuracy of cocoons,the method of manually identifying cocoon types has been difficult to meet the requirements of continuous production of enterprises.Therefore,how to improve the accuracy of cocoon classification and recognition has become the focus of enterprise attention and research.In order to improve the accuracy of cocoon classification and recognition,this paper mainly studies the cocoon classification and recognition technology based on deep learning.The research results are as follows:An image classification and recognition method of cocoons based on unsupervised region localization and feature fusion is proposed.The method firstly uses the pretrained Res Net50 network model to locate the cocoon area unsupervised,and then divides the cocoon area into two equal parts along the height direction of the image,and inputs the cocoon area and the divided image into the neural network for training and classification,and finally use image processing technology to reclassify the classified car cocoons and macular cocoons.The experimental results show that the method proposed in this chapter is 1.2% and 0.6% higher than the classification and recognition accuracy of VGGNet19 and Res Net50,respectively.A classification and recognition method of cocoons based on non-local attention mechanism and multi-scale receptive field fusion is proposed.This chapter introduces a multi-scale receptive field fusion module,a non-local module,and a channel attention module to improve the performance of the Res Net50 model in cocoon classification.A large number of ablation experiments prove that the method proposed in this chapter has a significant improvement in accuracy compared with the original Res Net50.The cocoon image recognition method based on efficientdet is realized.In this chapter,a data set suitable for simultaneous recognition of multiple cocoons is constructed by using the lower left corner filling algorithm,and then the data set is used to train the efficientdet network model.The experimental results show that the cocoon detection accuracy of Efficient Det D2 network is the highest and better than that of Yolo v5 s and SSD network models. |
