AN EXPLAINABLE DEEP LEARNING MODEL FOR CROP DISEASE DECTION

Abstract

In recent years, many studies have applied deep learning in artificial intelligence to assist in the detection and classification of plant diseases. However, these models, when applied in practice, often lack transparency and suffer from insufficient accuracy. In this paper, we use two explainable artificial intelligence (XAI) techniques to analyze how the model identifies diseases, providing explanations for predictions using the New Bangladesh Crop dataset, which is derived from the Plant Village dataset and focuses on key food crops. To evaluate the model's focus on diseased regions, we calculate the Intersection over Union (IoU) values for selected disease images from each crop. The experimental results guide the selection of appropriate XAI methods and help fine-tune the model for improved accuracy. We propose an enhanced VGG16 model with attention mechanisms, achieving relatively high accuracy and improved focus on diseased regions of plant leaves.