摘要
应用深度学习技术提出卷积神经网络(CNN)和Ghost海滩塑料制品检测模型。对于CNN海滩塑料制品检测模型,经仿真设计神经网络层数量为6,测试准确率为0.86;对于Ghost海滩塑料制品检测模型,经仿真设定合理的参数,测试准确率为0.92。在多场景海滩塑料垃圾检测任务中,Ghost凭借线性变换充分利用冗余塑料特征信息,具备较高检测准确率;与传统HOG检测模型相比,CNN和Ghost模型检测准确率分别高于HOG模型11%和17%。
CNN and Ghost beach plastic products detection models are proposed by using deep learning technology.For CNN Beach plastic products detection model,the number of neural network layers was 6 through simulation design,and the test accuracy was 0.86.For Ghost beach plastic products detection model,reasonable parameters were set through simulation,and the test accuracy was 0.92.In the multi scene beach plastic waste detection task,Ghost makes full use of the redundant plastic feature information by linear transformation,and gets a high detection accuracy.Compared with the traditional HOG detection model,CNN and Ghost model had higher detection accuracy than the HOG model by 11%and 17%,respectively.
作者
李杰
郭颖平
李晓敏
LI Jie;GUO Ying-ping;LI Xiao-min(Baoding Vocational and Technical College,Baoding 071051,China;Center For Hydrogeology and Environmental Geology Survey CGS,Baoding 071000,China)
出处
《塑料科技》
CAS
北大核心
2020年第4期78-82,共5页
Plastics Science and Technology