COMPARATIVE ANALYSIS OF DEEP LEARNING MODELS FOR PLANT DISEASE CLASSIFICATION IN DIFFERENT COLOR SPACES

Abstract

This study presents a comparative analysis of the impact of different color spaces on the accuracy of plant disease classification using convolutional neural networks. Three model architectures were trained in four color spaces – RGB, HSV, LUV, and LAB – resulting in a total of 12 models. The experiments were conducted on a dataset containing over 79,000 original images across 88 classes. Classification accuracy was evaluated on both training and test sets. The results indicate that classification performance depends not only on the model architecture but also on the chosen color space. In particular, LAB showed slight advantages in models with fewer parameters, while RGB consistently performed well in more complex models. HSV and LUV generally resulted in lower accuracy. These findings can contribute to improving plant disease diagnosis systems and may also be applicable in other fields such as metallurgy for enhancing the accuracy of defect detection on metal surface images.