Transparent Intelligent Vision for Black Sigatoka Detection

Black Sigatoka is a devastating fungal disease that affects banana plants worldwide. Early and accurate detection of the disease is essential for preventing its spread and reducing its impact. Convolutional Neural Networks (CNNs) have exhibited promise in detecting plant diseases, including Black Sigatoka. However, CNNs are often regarded as black boxes that lack transparency and interpretability, which limits their trustworthiness and applicability. This chapter introduces an approach that enhances transparency in Black Sigatoka detection, employing techniques such as saliency maps, occlusion analysis, Grad-CAM, and SHAP. Using ResNet-18 and a proprietary CNN model that has been thoroughly trained on a large dataset of images of banana leaves, our method provides insights that help the user better understand the CNN’s decision-making process. Extensive experiments confirm that our methodology outperforms the state-of-the-art methods used for Black Sigatoka detection at this time. This method not only guarantees clear and reliable results but also encourages moral AI practices in the field of diagnosing plant diseases.