Interpretability of Convolutional Neural Networks in Infrared Point Objects Classification

Deep learning has demonstrated remarkable advancements in the field of infrared point objects classification. However, the challenge lies in explaining the reasons behind their exceptional performance, which limits their application in high-reliability object recognition systems. To address this issue, this paper proposes a framework for interpreting convolutional neural network (CNN) models in the classification of infrared point objects. By leveraging the time sequence of infrared radiation (IR) emitted by point objects, we adopt the one-dimensional (1D) class activation map (CAM) technique to identify the positions within input subsequences that are relevant for classification. Experimental results demonstrate that our proposed method provides an explanation for how CNNs work in point objects classification. This not only enhances the reliability of CNN-based methods in point objects classification but also offers guidance for further improvement in model performance and interpretation of predictive results.