摘要
为了满足计算机视觉辅助下应用机器人进行扇贝自动分拣的实时性和鲁棒性要求,提出了一种基于神经网络的扇贝识别和分级方法。首先对图像进行灰度化处理,并用canny算子检测目标边界,然后用8-连通邻域追踪算法提取目标边界像素坐标,最后计算目标边界到中心点的平均距离及其绝对平均误差,并作为特征信息训练BP神经网络,实现对扇贝图像识别和分类。实验结果表明,该方法可以快速实现扇贝的自动识别和分级工作。
A scallop identification and classification method is proposed based on neural network to need the timeliness and robustness requirements of scallops sorting by computer vision aided robot automatically.Firstly,color image was converted into graying image,and object edges were detected by a canny operator on the binary image.Secondly,the pixel coordinates of object edges were extracted by eight neighborhood tracking method,and the mean and variance of the distance between image edge and image center were computed.Finally,BP neural network was trained to classifier of scallops using the mean and variance of the distance as classification feature.Experimental results indicated that the neural network classifier quickly and correctly completed the identification and gradation of scallops.
出处
《大连海洋大学学报》
CAS
CSCD
北大核心
2014年第1期70-74,共5页
Journal of Dalian Ocean University
基金
辽宁省博士启动基金资助项目(20071066)
辽宁省教育厅科研项目(L2010072)
关键词
扇贝识别
图像处理
特征提取
BP神经网络
scallop recognition
image processing
feature extraction
BP neural network
