The increasing demand for intelligent food inspection and post-harvest automation has accelerated the adoption of artificial intelligence and computer vision technologies in modern agricultural systems. However, many existing deep learning approaches for vegetable classification rely on computationally intensive architectures, limiting their deployment in resource-constrained environments commonly found in developing regions. To address this challenge, this study proposes LightVegNet, a lightweight and optimized convolutional neural network (CNN) framework for automated vegetable classification under practical agricultural and food processing conditions. LightVegNet integrates hierarchical feature extraction with selectively applied depthwise separable convolutions to achieve a balance between classification performance and computational efficiency. Experimental evaluation using 5-fold cross-validation demonstrated that the proposed framework achieved a mean classification accuracy of 99.31%, outperforming several widely used lightweight benchmark models. In addition, the proposed model maintains a compact architecture with only 2.13 million trainable parameters, offering substantial reductions in model complexity while preserving high predictive capability. The findings indicate that LightVegNet provides a promising solution for intelligent vegetable classification and may serve as a foundational component for future food inspection, grading, and post-harvest quality assessment systems in resource-constrained environments.