基于深度学习的果蔬识别系统

Fruit and Vegetable Identification System Based on Deep Learning

摘要: 在农产品分拣过程中，果蔬的分类大多还停留在传统的分类模式，由人工进行果蔬筛选，这不但增加了成本，还可能给劳动者带来不便的体验。得益于计算机视觉相关技术的不断发展，以机器代替人工劳动的运营模式逐渐成为可能。本文探讨了以Tensorflow深度学习框架为核心，利用OpenCV进行图像处理的果蔬识别系统。基于收集的果蔬图像数据，通过多次调参及训练获得了准确率较高的模型。在系统构建上对图像数据集的使用、模型参数调整、训练结果以及应用性能进行了可视化设计与分析，通过Web App实现并展示了图像识别、目标检测、语义分割、实时检测四大功能模块。

Abstract: In the processing of sorting agricultural products, the classification of fruits and vegetables mostly stays in the traditional classification mode. This mode increases the labor cost on the one hand and the inconvenience to the workers on the other hand. Thanks to the continuous development of computer vision technologies, replacing labor with machines has gradually become possible. The paper studies the model structure and principle used in the development of fruit and vegetable recognition systems. Tensorflow is applied as core deep learning framework and OpenCV is used as in image processing. Based on the collected images of fruits and vegetables, a model with high accuracy was obtained through multiple adjustments and training. Image datasets, model parameter adjustment, training results and application performance are visualized and analyzed in the system. Four functional modules of image recognition, object detection, semantic segmentation and real-time detection are realized in the Web APP.