Detecting Severity Levels of Cucumber Leaf Spot Disease using ResNext Deep Learning Model: A Digital Image Analysis Approach

Abstract

The fungal disease known as cucumber leaf spot (CLS) is capable of causing substantial damage to cucumber crops, leading to a decrease in production and quality. Early detection and management of the disease are critical for minimizing its impact on crop productivity. In this study, the authors developed a Residual Next-50 (ResNext-50) deep learning (DL) model based on 5 different severity levels for the multi-classification of CLS disease. The work collected a dataset of 50,000 digital images of cucumber leaves from multiple sources, including local farmers and agricultural research institutions. The dataset was segmented into training, validation, and testing groups, each of which contained 7,500 pictures. The training set contained 35,000 photos. The proposed model achieved an overall accuracy of 97.81% on the testing set, outperforming several baseline models, including U-Net, YOLO v5, and KNN. The confusion matrix revealed that the model was most accurate in identifying cucumber leaves with very low severity of the disease (Level 1) and very high severity of the disease (Level 5) while having lower precision values for cucumber leaves with moderate severity of the disease (Level 3). The suggested study demonstrates the potential of DL models for the accurate and efficient classification of CLS disease based on severity levels, providing valuable insights for the early detection and management of the disease. The insight that was gleaned from this research could also contribute to the development of methods that are more efficient in the prevention and management of CLS illness, ultimately improving the sustainability and productivity of cucumber cultivation.