摘要
目前鱼类种类繁多,人工识别面临着劳动力老年化和短缺的挑战,为解决人工识别存在的问题,采用一种轻量级卷积神经网络MobileNetV2,在30种鱼类数据集的基础上,对神经网络进行迁移学习,得到paddle分类模型,利用模型转换工具,得到可以在K210模块上运行的模型。实验结果表明,该系统可以以较高的准确率完成鱼类的识别,实时性较好,功耗低。
As there are varieties of fish at present,the artificial fish recognition is facing the challenge of labor aging and shortage.In order to solve this problem,a lightweight convolutional neural network named MobileNetV2 is introduced.On the basis of 30 kinds of fish data sets,it can obtain the paddle classification model by using the transfer learning of the neural network.Then a model that can run on k210 module is available with the help of model conversion tool.The experimental results show that the system has high accuracy in fish recognition,with good real-time performance and low power consumption.
作者
林志谋
LIN Zhimou(Department of Mechanical&Electrical Engineering,Xiamen Ocean Vacational College,Xiamen Fujian 361012,China)
出处
《顺德职业技术学院学报》
2023年第1期26-30,共5页
Journal of Shunde Polytechnic
基金
福建省教育厅“智慧渔业应用技术协同创新中心”鱼类智能识别项目(XTZX-ZHYY-1917)
厦门市科技局自然科学基金项目(3502Z20227430)。
关键词
鱼类
智能识别
神经网络
百度飞桨
fish
intelligent recognition
neural network
PaddlePaddle
