Enhanced CNN for Fruit Disease Detection and Grading Classification Using SSDAE-SVM for Postharvest Fruits

In the realm of agriculture, leveraging image processing has become pivotal for robust image analysis, especially in detecting fruit diseases. However, existing techniques in this domain often limit inputs to fixed sizes without reshaping images before neural network (NN) input, complicating disease detection and compromising image resolution, thereby escalating postharvest losses. To address this, an innovative approach has been developed a unique enhanced convolutional NN (CNN) employing spatial pyramid pooling (SPP) and adaptive momentum BP that integrates the best finite impulse response (FIR) filter for preprocessing. This method aims to reorganize the fruit detection process while maintaining high image resolution. The CNN, with its SPP, utilizes convolutional layers to extract diverse features encompassing color, shape, texture, and surface attributes crucial for accurate disease detection. Furthermore, efficient fruit grading is essential to combat issues such as poor product quality, slow grading speeds, and accuracy concerns, all contributing to postharvest losses. In response, a novel integrated stacked sparse denoising autoencoder–support vector machine (SSDAE-SVM) approach, coupled with dropout mechanisms, has been proposed to streamline fruit grading and mitigate postharvest losses. The strategic use of dropout layers mitigates overfitting and information loss during feature extraction, while the SVM classifier, serving as the output layer, ensures accurate fruit grading, thereby curbing postharvest losses. Consequently, this proposed method not only simplifies disease detection and grading processes but also enhances quality, accuracy, reliability, and speed. The model's performance surpasses previous disease prediction models, exhibiting an impressive accuracy of 97.25%, a minimal prediction error of 0.15, a high specificity of 95.62%, an ${F}1$ -score of 98.81%, and a remarkable recall rate of 98.98%.