摘要 为了解决安格斯肉牛图像目标与背景颜色相近导致常用方法无法准确提取牛只轮廓及识别牛只姿态的问题,试验采用最近邻(K-nearest neighbor,KNN)算法分割图像,计算Hu七阶不变矩,分析牛体姿态,在此基础上改进算法以进一步识别图像中的区域叠加部分。结果表明:改进后的算法弱化了K值取值大小的影响,能准确地提取出目标图像,基于Hu不变矩算法识别站、卧姿态的正确率均超过90%。说明采用的KNN算法和Hu七阶不变矩方法能完整地分割出复杂背景下的安格斯肉牛图像目标,正确分析出牛只姿态。 In order to solve the problem that target profile and recognition posture of Angus cattle image cannot be accurately extracted by conventional methods due to the similarity between the image target and the background,the K-nearest neighbor(KNN)was used in the experiment to segment the image,calculate the seven invariant moments of Hu,and then analyze the posture of the cattle body.On the basis,the improved algorithm was used to further identify the superposition in the image.The results showed that the improved algorithm weakened the influence of K value size,and could extract the target image accurately.based on Hu’s invariant moment algorithm,the accuracy of both standing and lying posture recognition was over 90%.The results indicated that the target extraction and posture identification of Angus cattle image could be segmented completely under complex background by using KNN and Hu’s seven invariant moments.
机构地区 宁夏农林科学院农业经济与信息技术研究所
出处 《黑龙江畜牧兽医》 CAS 北大核心 2021年第6期42-45,49,144-146,共8页 Heilongjiang Animal Science And veterinary Medicine
基金 宁夏农林科学院全产业链创新示范项目“肉牛健康养殖监测预警系统的研究与示范应用”(QCYL-2018-1104)。
关键词 安格斯肉牛 最近邻算法 轮廓提取 HU不变矩 欧氏距离 姿态识别 Angus cattle K-nearest neighbor contour extraction Hu’s invariant moments euclidean distance posture recognition
