AI-driven lightweight CNN model for sustainable vegetable classification in smart food systems

Abstract

The increasing need for sustainable and environmentally friendly technological solutions in agriculture requires AI approaches that are both precise and computationally efficient. In this context, this study introduces a lightweight AI-driven convolutional neural network (CNN) that is designed for sustainable vegetable classification within intelligent food systems. The model is characterized by a compact architecture that incorporates convolutional layers for spatial feature extraction and Squeeze-and-Excitation (SE) blocks for adaptive channel-wise attention. It achieves a high classification accuracy of 98% while maintaining a minimal computational burden with only 0.39 million parameters and 15.63 GFLOPs. Compared to deeper models such as ResNet18 and GoogLeNet, the proposed model demonstrates superior efficiency and faster inference. Its low power consumption and real-time processing capabilities make it particularly suitable for implementation in resource-limited, environmentally friendly agricultural settings.