摘 要：近年来，我国柑橘产业迅猛发展，柑橘的分拣需要从传统的人工分拣进入智能分拣阶段。该系统通过使用百度Easy DL 开放平台中图像识别模型进行柑橘的大小分类，完成训练后发布到公有云，开发一个 Android APP 调用云接口实现拍照识别爱媛、青枧两种柑橘的分类。该系统的核心是柑橘识别模型的机器训练，通过采集柑橘照片、进行分类标注，将数据提交到模型进行机器学习，根据训练结果增加训练数据再次训练。经过多次迭代，最终达到目标识别精度。
关键词：图像识别；机器学习；柑橘分类
DOI:10.19850/j.cnki.2096-4706.2021.11.023
基金项目：成都农业科技职业学院科研项目 （20ZR210）
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2021）11-0086-03
Design and Implementation of Citrus Classification Software Based on Image Recognition
LEI Jing
(School of Electromechanical and Electrical Information, Chengdu Agricultural College, Chengdu 611130, China)
Abstract: In recent years, with the rapid development of citrus industry in China, citrus sorting needs to change from traditional manual sorting stage to intelligent sorting stage. The system uses the image recognition model in Baidu Easy DL open platform to classify the size of citrus, releases it to the public clouds after finishing training, and calls the cloud interface through a developed Android APP to realize the classification of two kinds of citrus, namely, Aiyuan and Qingjian by taking photos. The core of this system is the machine training of citrus recognition model. Through collecting citrus photos, classifying and labeling, the data is submitted to the model for machine learning, and the training data is added according to the training results for training again. After several iterations, the target recognition accuracy is achieved finally.
Keywords: image recognition; machine learning; citrus classification
参考文献：
[1] 田楠坤 . 全国及四川省柑橘生产现状及未来发展预测 [J]. 粮食科技与经济，2019，44（7）：129-131.
[2] 倪云峰，叶健，樊娇娇 . 基于图像识别的水果分拣系统 [J]. 江苏农业科学，2021，49（10）：170-176.
[3] 王颖 . 打造产业集群，四川柑橘破解“盛产”危机 [J]. 营 销界，2020（32）：26-30.
[4] EasyDL 零门槛 AI 开发平台 . 什么是 EasyDL [EB/OL]. [2021-04-11].https://ai.baidu.com/ai-doc/EASYDL/pk38n 3odv.
[5] 张悦湘 . 浅谈水果检测及分拣研究进展 [J]. 南方农机， 2020，51（15）：73-75+77.
作者简介：雷静（1981—），女，汉族，四川成都人，副教授， 工程硕士，研究方向：计算机应用。