Improving transparency in machine learning for agriculture: a case study on mushroom classification using CNN and Grad-CAM

Machine learning (ML) is a powerful tool for precision agriculture (PA) in general. However, its widespread adoption is constrained by lacking transparency. To address this gap, this study integrates ML with Explainable AI (XAI) methods to enhance both the predictive performance and interpretability of the respective models. We provide a review of ML and XAI techniques suitable for PA, with a specific focus on Convolutional Neural Networks (CNNs) and Gradient-weighted Class Activation Mapping (Grad-CAM). Furthermore, we conducted a case study on mushroom hyperspectral image classification. A 2D-CNN model was developed, achieving average accuracy 98.04%, precision 98.96%, recall 96.30%, and F1-score 97.51%. Grad-CAM was then applied to identify the critical pixels that informed the model’s classifications. The results demonstrate that the integration of CNNs with Grad-CAM not only achieves high predictive accuracy on complex agricultural data but also enhances model interpretability.