Classification of Trash Types in Cotton Based on Deep Learning

A classification method of trash types in cotton using depth-wise separable convolutions to enrich features in deep neural networks is proposed. The method based on Vgg16 what is a typical neural network adopts depth-wise separable convolutions and deals with the output feature of every layer to rich feature much further. Then, the pointwise convolution is used to connect the features. Finally, the fully connected layer combines the whole features to classify trash types. The experiment results show that this method can improve the classification accuracy of trash types in cotton.