Enhancing Remote Sensing Image Classification with DWT and CSG-CAM for Visual Explanations

To develop a method for visual explanations in picture classification that incorporates detailed spatial information, combining channel saliency analysis with wavelet-based area identification. Utilizes a Convolutional Neural Network (CNN) model for image classification tasks. This approach focuses on end-to-end feature extraction and classification using standard CNN architectures with a sample of 32 samples. Introduces Discrete Wavelet Transform (DWT) to map distinct spatial regions and identify critical channels with a sample of 51 samples. These wavelet-transformed maps are combined with salient object identification results, emphasizing important regions for improved accuracy. The method achieved an accuracy above 90%, providing practical visual representations that enhance interpretability and reliability in image classification with the significance value below 0.05.The proposed approach DWT improves the accuracy and reliability of categorizing results and efficient than the existing CNN algorithm.