Machine-Learning-Image-based-Fruits-and-Vegetables-Recognition-with-calories-

ABSTRACT

This project introduces a lightweight deep learning model for fruit and vegetable recognition, combining a modified MobileNetV2 architecture with an attention module. The model begins by extracting convolutional features to capture essential object-based information, followed by an attention mechanism that enhances semantic understanding. By integrating these modules, the model effectively balances high-level object representation with nuanced semantic details crucial for accurate classification. Utilizing transfer learning from a pre-trained MobileNetV2 model optimizes training efficiency and adaptation to specific fruit datasets. Evaluation across three benchmark datasets demonstrates that the proposed model surpasses four recent deep learning methods in classification accuracy while significantly reducing the number of trainable parameters. This makes it well-suited for applications in industries related to fruit cultivation, retail, and processing, where automatic fruit identification are valuable for operational efficiency and quality control.