Improving the Classification of Fruit Images: Denoising Techniques for Better Dataset Quality

The performance of machine learning models in the computer vision for the task of picture classification has greatly affected the quality of image datasets. Since noise from images can reduce the accuracy of model, noise reduction is a key preprocessing step. To clean the noisier images from a mixed fruit collection, this work investigates the application of enhanced filtering techniques, including PDTMF, CWF, GBF, and SIGF. This study aims to determine how different filters enhance image quality and how well they preserve the essential features for accurate fruit identification. The ability of each method is evaluated by applying standard image quality measures such as the Structural Similarity Index (SSIM), the Peak Signal-to-Noise Ratio (PSNR), and visual inspection. It has been found that though GBF and SIGF are good in removing some types of noise, i.e., Gaussian and salt and pepper noise respectively, PDMTF and CWF are the best in removing noise with least information loss about the image. So, this paper contributes to developing more reliable picture preprocessing techniques for fruit classification systems to enhance the accuracy and reliability of the automatic fruit recognition systems.