基于Squeezenet模型的蛇类图像识别  

Snakes Image Recognition Based on Squeezenet

作　　者：鲁元平 罗志聪[1] LU Yuan-ping;LUO Zhi-cong(College of Mechanical and Electrical Engineering Fujian Agriculture and Forestry University,Fuzhou 350002,China)

机构地区：[1]福建农林大学机电工程学院,福建福州350002

出　　处：《电脑知识与技术》2021年第7期12-14,共3页Computer Knowledge and Technology

基　　金：福建省自然科学基金项目(项目编号:2018J01533)。

摘　　要：随着人类活动范围的不断扩张,生态环境的恶化,蛇类的生存环境受到威胁。为实现对蛇类的准确识别,辅助蛇类研究者对蛇类的识别与保护,通过对现有的解决方案进行分析,提出采用迁移学习方法对10种蛇类进行训练识别。首先对现有数据集进行数据增强,从而让训练模型更好地收敛,采用基于Squeezenet模型具有更好的识别率,识别率达到了85.71%,效果良好。并对训练好的模型进行加载,网络模型识别耗时2.69s,有着更好实时性,同时还能保证识别精度,节省硬件资源,提升了模型的可移植性。With the continuous expansion of human activities,the deterioration of the ecological environment,the living environ⁃ment of snakes is threatened.In order to realize the accurate identification of snakes and assist the snake researchers in the identifi⁃cation and protection of snakes,this paper proposes to adopt the transfer learning method to train and identify 10 kinds of snakes by analyzing the existing solutions.Firstly,the existing data set was enhanced to promote better convergence of the training model.With the adoption of the Squeezenet model,the recognition rate reached 85.71%,with good effect.It takes 2.69s to identify the net⁃work model,which ensures better real-time performance,ensures recognition accuracy,saves hardware resources,and improves the portability of the model.

关 键 词：迁移学习 Squeezenet模型 蛇类 图像识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]