Enhancing Apple Quality Identification: A Multi-Layer Visual Feature Fusion Approach with Convolutional Neural Networks

This research uses real-world images with complicated disturbance information (the backdrop was identical to the apples' surfaces) to identify and categorize apple quality. This research introduces a new model for the quick and accurate evaluation of apple quality based on convolutional neural networks (CNNs). The proposed model captures detailed, unique, and valuable image features for detection and classification. In contrast to the deep features, complex interpretation needs and large training sample requirements, the handcrafted features used in this work are mainly produced intuitively. A new feature fusion approach called multi-layer visual feature fusion (MLVSF) is created for better classification of apple images. The MLVSF model, which incorporates deep and handcrafted components obtained from bag-of-visual words, local binary pattern variations, and CNN, can improve the discriminatory power of features used for apple quality identification. With its ability to improve CNN and achieve more accurate classification—resulting in a classification accuracy of 98.93%—MLVSF beats other modern approaches when evaluated on apple image datasets.