Robust Classification of Red Chili Plant Leaves Using Smartphone Camera Data and ResNet Model in Noisy Environments

Classification of yellow leaves, curly leaves and leaf spots on red chili plants is an important task in monitoring plant health. However, real-world scenarios often involve challenges such as environmental variations and noise from different leaf types, which can hinder accurate classification. This study aims to develop a robust classification model for identifying the three mentioned leaf types under dynamic and noisy conditions using smartphone camera data. Data collection involved capturing leaf images with a smartphone camera at a fixed distance. The images were intentionally taken with the inclusion of noise from surrounding leaves. The collected data were then preprocessed, including resizing and adjusting image format and resolution. Three scenarios were evaluated: default ResNet, ResNet with Grid Search-tuned model, and ResNet with Random Search-tuned model. The models were evaluated using metrics such as Accuracy, Precision, Recall, and F1-Score. The default ResNet scenario demonstrated moderate performance, while the ResNet model with Grid Search-tuned parameters showed the best overall performance with improved Precision and F1-Score. The model's ability to classify the three classes was highlighted, considering the trade-off between Precision and Recall. Collecting data with intentional noise from different leaf types simulates real-world conditions, aiding in the development of a robust classification model. The ResNet model tuned through Grid Search yielded the best performance. Further validation on larger datasets and improving adaptability to environmental variations is advised. The research contributes to effective plant disease detection in agricultural practices, leveraging smartphone cameras for efficient field observations and early problem identification.