摘要
为提取高效的植物形态学特征,提升在图像中植物的识别准确率,结合植物的宏观整体轮廓和微观局部纹理特征,提出融合式双特征神经网络(HDF-CNN)进行植物识别。采用并行双通道方案,充分利用密集残差网络增强植物器官细节纹理描述,采用另一个大卷积核CNN提取植物全局轮廓特征,将宏观和微观的特征融合后得到一个全面的植物特征。实验结果表明,与常见的图像识别方法相比,该方法能提取到精准高效的植物图像特征,取得了更高的识别准确率。
To extract efficient morphological features of plants and improve the recognition accuracy of plants in the image,a fusion dual feature neural network(HDF-CNN)was proposed to recognize plants by combining the macro overall contour and micro local texture features of plants.The network adopted a parallel dual channel scheme,which made full use of the dense resi-dual network to enhance the detailed texture description of plant organs.Another CNN with large convolution kernels was employed to extract the global contour feature of plants.The macro and micro features were combined to obtain a comprehensive plant feature.Experimental results show that compared with common image recognition methods,the proposed method can extract accurate and efficient plant image features,and a higher recognition accuracy rate is achieved.
作者
左羽
徐文博
吴恋
ZUO Yu;XU Wen-bo;WU Lian(School of Mathematics and Big Data,Guizhou Education University,Guiyang 550018,China;Big Data Science and Intelligent Engineering Research Institute,Guizhou Education University,Guiyang 550018,China)
出处
《计算机工程与设计》
北大核心
2021年第6期1706-1712,共7页
Computer Engineering and Design
基金
贵州省科技计划基金项目(黔科合基础[2018]1121)
贵州省科技厅国家科技部和国家自然科学基金奖励补助基金项目(黔科合平台人才[2017]5790-09)
贵州师范学院一流大学建设基金项目(贵师院发[2018]100号)
贵州省科学技术基金计划基金项目(黔科合基础[2016]1116)。
关键词
计算机视觉
植物识别
植物特征
特征融合
混合模型
computer vision
plant recognition
plant feature
feature fusion
hybrid model
