CLayerCAM: A Model for Explaining Aerial Scene Classification using Convolutional Neural Network
Year: 2025
Type: conference-paper
Abstract: Motivated by the pressing need to achieve transparent decision-making of the deep network models in safety-critical applications, such as landslide detection, disaster management, and so on, this research aims at investigating explainable artificial intelligence (XAI) to improve the interpretability as well as trustworthiness of convolutional neural networks (CNNs) for scene-level classification of aerial images. Our research contributes towards advancing the field of XAI by proposing a channel-wise confidence-driven class activation mapping (CAM) approach which is also applicable on any layer/stage of a CNN-based prediction model. The proposed confidence-driven approach, termed as CLayerCAM, integrates the benefits of existing Score-CAM and LayerCAM methods so as to achieve improved interpretability on the decisions made by the CNN-based scene classifier, and thereby, enhances user trust in the model’s predictions. Extensive experimentation on benchmark aerial scene datasets and evaluation with respect to the remove and debias (ROAD) metric demonstrates the efficacy and supremacy of our proposed CLayerCAM over the state-of-the-arts.
Source: Procedia Computer Science
Authors: Yogeshkumar Sant, Monidipa Das
Institution: Indian Institute of Technology Dhanbad
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W4410258449