Research on Identification Method of Wild Mushrooms Based on Compact Bilinear Network
Wild mushrooms are favored by people because of their delicious taste and high nutritional value.In recent years,poisoning deaths due to eating wild mushrooms have occurred frequently,so deep learning is used to identify wild mushrooms.In order to solve this problem,an improved Efficient Compact Bilinear Convolutional Neural Network(Efficient-CBCNN)is proposed from the perspective of fine granularity.The bilinear network framework is adopted.Firstly,the classification layer of EfficientNetV2 model is removed as a branch of the bilinear network to extract features;Secondly,the Efficient Multi-scale Attention(EMA)mechanism with better performance is introduced to improve EfficientNetV2,which can maintain the performance while reducing the computational load.Then the branch EfficientNetV2 structure is simplified to reduce the complexity of the structure and computing overhead.Then,the Compact Bilinear Pooling(CBP)is accessed to pool the features output from the two branches,capture the higher-order interactive information between the two feature maps,and enhance the expression ability of features.Finally,connect the user-defined full connection layer for classification.The experimental results show that the recognition accuracy of Efficient-CBCNN model has been greatly improved,achieving 98.49％accuracy.Compared with the four models VGG16,ResNet50,ShufflenetV2,and BCNN(original),the accuracy of the proposed model has been improved by 10.17％,6.19％,12.82％,and 5.33％respectively,and the number of parameters is less than that of BCNN(original),and the training speed is faster.
identification of wild mushroomEfficientNetV2EMAcompact poolingfine grained
国家科技期刊平台
NSTL
万方数据
