Dried shiitake mushroom grade recognition based on improved ResNet18
To solve the problems of complexity and low recognition accuracy of dried shiitake mushroom grade recognition technology,a method of dried shiitake mushroom grade recognition based on residual neural network ResNet18 is proposed.Firstly,the 7×7 convolutional layer of Stem in the traditional ResNet18 is replaced by three 3×3 convolutional layers in series,which ensures that the computational amount is further reduced while the sensory field remains unchanged.Secondly,to address the problem of insufficient linear and nonlinear transformations in the residual block,fused asymmetric convolution and h-swish activation function are introduced,which increases the complexity of the model and enables it to carry out a deeper level of feature learning.Finally,an efficient channel attention mechanism is introduced into the ResNet18 backbone network to strengthen the ability of the model to extract features.The experimental results show that the improved ResNet18 network model has an accuracy of 97.04%,which is 4.81%higher compared to the ResNet18 network modeling method,and outperforms VGG16,MobileNetV2,DenseNet121,ResNet34 and other network model methods,which can improve the recognition accuracy of dried shiitake mushroom grades,and the detection time of a single image is 5.91 ms,which is useful for grade recognition in the intelligent sorting process of dried shiitake mushrooms.
NETL
NSTL
万方数据
