Automatic Fruit and Vegetable Classification Using Deep Learning and Computer Vision: An Approach for Intelligent Weighing Systems

Automatic fruit and vegetable recognition is an essential component of intelligent weighing systems in supermarkets and self-service stores. These systems must operate under strict latency and hardware constraints while maintaining high classification accuracy under varying lighting, backgrounds, and occlusions. This work investigates compact deep-learning classifiers using the Fruits and Vegetables Detection for YOLOv4 dataset. We build a YOLOv8s-based pipeline trained on high-resolution RGB images labeled by product type and the presence of plastic wrapping (bagged vs. non-bagged). First, we merge bag information into eight semantic classes (chilli, lemon, banana, apple, tomato, grapes, raspberry, blackberries), achieving 100% accuracy. We then refine the taxonomy to fourteen classes by splitting six categories into with-bag and without-bag variants, while keeping raspberries and blackberries only in the non-bag condition; the model again reaches 100% accuracy. We report a mean latency of 193 ms per image and a model size of 9.81 MB, and show that predictions remain stable under strong brightness changes and partial occlusions. Finally, we discuss overfitting risks and outline cross-dataset validation and embedded deployment as next steps.