Research on ResNet101 Network Chemical Reagent Label Image Classification Based on Transfer Learning
Zhengguang Xu, Kuo Sun, Junying Mao
Abstract
Chemical reagent label image classification research plays an important role in dealing with massive reagent classification tasks in chemical laboratories. To solve the lack of performance of traditional image processing methods in image classification and misclassification caused by purely manual classification, as well as there are too few image training data and a lot of manual annotations. This paper proposes a ResNetl01 network classification method based on transfer learning. Under the PyTorch deep learning framework, using ResNetl01 as the basic network framework, and then introducing the transfer learning method, first use the ResNetl01 network parameters trained in the ImageNet dataset to initialize the convolution layer parameters of the chemical reagent label classification model, and then use the chemical reagent label image data train the model, fine-tune the model parameters, and finally get the classification model. The experimental results show that the classification method can obtain a good accuracy on the chemical reagent label image data set.