Key points are not available for this paper at this time.
For the count of cocoons, the image is segmented by kmeans method in the first place, then the distance transformation method is adopted to obtain the distance gray image, and the cocoons that are adhered to each other are separated by morphological operation to obtain the image of unconjugated cocoon. Finally, the cocoon counting is realized by traversing the connected domain. For cocoon classification, a fine tuned AlexNet neural network is used to classify cocoons, and batch normalization is used to replace local response normalization in conv1 and conv2.The data in the network is normalized to a normal distribution with variance of 1. This method can not only accelerate the training speed of the network, but also slow down the gradient disappearance problem in the training process, and improve the generalization ability of the network. Finally, higher recognition accuracy rate is obtained than the AlexNet network without improvement.
Wang et al. (Sat,) studied this question.