Detecting fungi-affected multi-crop disease on heterogeneous region dataset using modified ResNeXt approach

Crop diseases pose significant threats to agriculture, impacting crop production. Biotic factors contribute to various diseases, including fungal, bacterial, and viral infections. Recent advancements in deep learning present a novel approach to the detection and recognition of these crop diseases. While considerable research has focused on identifying and recognizing crop diseases, fungal disease-affected crops have received relatively less attention and also detecting disease on different region datasets. This paper is about spotting fungal diseases in crops across different regions with diverse climates. It emphasizes the need for tailored detection methods, addressing the risk of mycotoxin production by fungi, which can harm both humans and animals. Detecting fungal diseases in apple, guava, and custard apple crops such as spot, scab, rust, rot, leaf spot, and insect ate. In the proposed work, the modified ResNeXt variant of the convolution neural network (CNN) technique was employed to predict 3 major crop classes of fungal disease. Initially, using Inception-v7 and ResNet for fungal disease in crops did not yield satisfactory results. A modified ResNeXt CNN model was proposed, showing improved fungal disease prediction. The novel model underwent a comparison with established methodologies. The suggested model draws upon a benchmark dataset consisting of 14,408 images capturing fungal diseases, categorized into three distinct classes: apple, custard apple, and guava. Experimental outcomes show that the proposed mutated ResNeXt model outperformed the state-of-the-art approaches. The model achieved 98.92% accuracy and high performance across recall, precision, and F1-score (above 99%) for the benchmark dataset, which gained encouragement and was comparable with the state-of-the-art approach.