Intelligent Strategy by Deep Learning for Thyroid Case Images Classification
Li Su · 2020
The sparse network structure of traditional convolutional neural network cannot keep the high efficiency of dense computation of fully connected network and the inaccuracy of classification results or slow convergence rate caused by low utilization rate of convolution features in the experimental process. This paper proposes an effective feature extraction scheme by increasing the number of convolution kernels. The strategy can reasonably expand the data set to make the model converge faster. The influence of different activation functions on the experiment is compared, and the images are classified by Softmax method. Finally, the improved model is used to classify the normal and pathological thyroid tissue images. The classification result of the model is better, and the average classification accuracy is 96.6%.