Identification of Downy Mildew Disease Stages in Salad Cucumber Using Machine Learning

Salad Cucumber (SC) is a commercially important crop grown in Sri Lanka, which spans a life cycle about 120 days. Downy mildew (DM) is a critical disease affecting different stages of SC life cycle, which significantly reduces the yield and shortens the growing season. Early detection and management of DM is essential for maintaining crop health and reducing economic losses. DM has different stages in its progression, and the early detection of those stages is important to adopt measures to treat and control. Often, early signs of DM go unnoticed, leading to detection only after severe damage to SC cultivation. Although extensive research exists on identifying DM in SC, studies focusing on stage-wise disease identification remain limited. During the study, a stage wise image dataset for DM was collected from selected tunnel environments in Monaragala and Matara districts. After pre-processing, the dataset was then compared under different experimental data setups using MobileNetV2, EfficientNetB0, ResNet50, VGG16, DenseNet121 and a customized Convolutional Neural Network (CNN). The customized CNN outperformed other models, achieving the highest classification accuracy of 93%, demonstrating its effectiveness in recognizing different stages of DM in SC. When integrated with relevant agronomic information, customized CNN can support early detection, prevention, and timely application of remedies for DM in SC where it shows strong potential as a practical solution for cultivators. This highlights how machine learning (ML) can assist disease detection and monitoring, thereby supporting precision agriculture applications and improving decision-making in crop production.