A Efficient Multi-label Cherry Classification Approach Based on Convolution Neural Network

Advancements in smart agriculture are increasingly pivotal for crop quality control, with computer vision playing a critical role in identifying defects. In this paper, we address the challenge of efficiently detecting cherry defects by developing a multi-label classification model using quantized convolutional neural networks. Our model innovatively fuses visible and infrared imagery to enhance feature extraction, a vital step in accurately identifying a wide array of defects. We applied a quantization strategy to the neural network's parameters, striking a balance between computational efficiency and the high accuracy required for practical agricultural applications. Our experimental results, which surpass existing benchmarks across key performance metrics, demonstrate that our model not only achieves an outstanding mean Average Precision but also excels in precision, recall, and F1 score. This research underscores the model's potential for deployment in on-field scenarios, offering a scalable and effective solution for automated cherry grading and quality assessment.