摘要
针对在野生濒危动物保护中,存在着对保护动物监测难度大、准确率低等问题。本文利用PyTorch框架,以VGG16为基础的SSD算法,提出了一种对濒危物种的动物识别与检测方法。在基础网络中,加入了注意力机制,并且在训练过程中使用迁移学习和数据扩增方法,提升了检测速度与精度。通过对数据集的测试,此方法不仅准确率达到了96.6%,而且其检测速度与模型大小符合预期,适用于对濒危动物的检测识别。
In the protection of wild endangered animals,there are problems such as difficulty in monitoring protected animals and low accuracy.Based on the PyTorch framework,a method for identifying and detecting endangered species of animals based on the VGG16 SSD algorithm is proposed.In the basic network,an attention mechanism is added,and migration learning and data amplification methods are used in the training process to improve the detection speed and accuracy.The test on the data set shows that the accuracy of this method reaches 96.6%.And its detection speed and model size are in line with expectations,which is suitable for detection and identification of endangered animals.
作者
蒋飞
王霄
JIANGFei;WANG Xiao(School of Electrical Engineering,Guizhou University,Guiyang 550000,China)
出处
《智能计算机与应用》
2020年第8期26-32,共7页
Intelligent Computer and Applications
基金
国家自然科学基金(61861007,61640014)
贵州省工业攻关项目(黔科合支撑[2019]2152)
黔科合人才团队((2015)4014)
物联网理论与应用案例库(KCALK201708)
自动化专业卓越工程师计划(ZYS 2015004)
关键词
濒危动物检测
迁移学习
SSD算法
深度学习
endangered animal detection
migration learning
SSD algorithm
deep learning
