PADDY LEAF DISEASE CLASSIFICATION USING ADVANCED DEEP LEARNING MODELS WITH NOISE REMOVAL APPROACHES

Abstract

Effective noise removal is essential for enhancing the quality of paddy leaf images used in disease prediction and analysis. This study compares the performance of three widely used filters: the Median Filter, the Bilateral Filter, and the Non-Local Means (NLM) Filter. The evaluation focuses on three key metrics: Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Measure (SSIM). Experimental results demonstrate that the NLM Filter consistently outperforms the Median and Bilateral Filters across all metrics. The Median Filter, although effective at removing salt-and-pepper noise, introduces slight blurring and fails to preserve intricate image details. The Bilateral Filter balances noise reduction and edge preservation but is less effective for complex noise patterns. The NLM Filter, leveraging its ability to identify and average similar patches throughout the image, achieves the lowest MSE, highest PSNR, and highest SSIM values, preserving both texture and structural details. This study highlights the superiority of the NLM Filter for noise removal in paddy leaf images, making it the preferred choice for preprocessing tasks in agricultural image analysis. After noise removal, classification was conducted using ResNet, InceptionNet, and EfficientNet. The combination of noise removal and classification methods—NLM + ResNet, NLM + InceptionNet, and NLM + EfficientNet—was analyzed. Among these, NLM + ResNet produced better results in terms of accuracy rate, precision, recall, and F1 score. The findings can guide researchers and practitioners in selecting optimal filters and DL models for improving image quality in automated disease detection systems.