ResNet-based image recognition model: interpretability and visualization research

In view of the fact that fruit image classification involves many different kinds of fruits, each fruit also has various forms and colors, so in the training of experimental models, many black box models that are difficult to understand are often involved. In order to improve the interpretability of the fruit image classification model and the understanding of the classification mechanism, this paper uses the deep learning model for experiments, and analyzes and discusses the generated CAM (Class Activation Map) heat map. The experimental results show that the fruit image recognition model based on ResNet (Residual Network) has high accuracy, and the generated CAM heat map can effectively explain the prediction process of the model. This paper provides an effective interpretability analysis method for deep learning models, which provides support for research and application in related fields.