A Robust Image Enhancement System Designed to Improve Plant Leaf Images for Machine Learning based Disease Detection

Abstract

Plant disease detection is critical for ensuring agricultural productivity and food security, yet the performance of machine learning models is often limited by the quality of input images. This study presents a robust image enhancement system designed to improve plant leaf images for machine learning-based disease detection. The system integrates three complementary techniques such as Non-Local Means (NLM) filtering which was used for noise reduction, then Wiener filtering used for image deblurring and Contrast Limited Adaptive Histogram Equalization (CLAHE) which was finally used for contrast enhancement and haze removal. Plant leaf images were collected from three farms in Uzu-Uwani, Enugu State, Nigeria and they underwent preprocessing steps including resizing, normalization and class balancing using SMOTE. Then the enhanced images were evaluated using a YOLOv5-based plant disease detection model for cassava and maize leaves. The results from the system implementation demonstrate that images processed with the proposed enhancement techniques significantly improved disease detection accuracy, thereby enabling the identification of multiple disease types that were otherwise missed in raw images. The findings highlight the importance of image enhancement in agricultural machine learning pipelines, providing a practical tool for researchers, agronomists, and farmers to improve disease monitoring and crop management.