机器视觉奶牛姿态识别研究+源代码
时间:2019-04-27 15:01 来源:毕业论文 作者:毕业论文 点击:次
| 摘要：为实现数字化养殖中奶牛异常姿态的自动检测和预警，本文以奶牛场2天的监控视频为研究对象，从视频中截取奶牛站立、躺卧两大类共八小类姿态图像，建立了姿态图像库后，对奶牛图像进行基于标记控制的分水岭分割和支持向量机分割实现目标提取。然后，提取奶牛对象的Hu不变矩等特征。最后，分别使用支持向量机、有监督学习神经网络GRNN和PNN、无监督学习神经网络SOFM对奶牛姿态进行识别。实验中，对图像库中74张奶牛姿态图像进行识别，识别正确62张，识别率为83.78%，达到很好的识别效果。34933 毕业论文关键词：奶牛姿态识别；图像处理；特征提取；支持向量机；神经网络 The Cow Posture Recognition Research Based on Machine Vision Abstract: In order to detect the cow abnormal posture and warn automatically for digital cultivation, the monitoring video of 2 days in dairy farm was taken as the research object in this essay. Images of standing and lying cows including 8 types of postures were captured from the video and the cow posture image library was established. Firstly, object of cow was segmented from image by two ways which are Marker-Controlled Watershed segmentation method and Support Vector Machine segmentation method. Then, the Hu invariant moments were extracted for the cow object. Finally, the cow postures were classified with three methods which are Support Vector Machine, supervised learning neural network GRNN and PNN, unsupervised learning neural network SOFM. In the experiment, 74 cow posture images were identified in the image library, and 62 correct ones were identified, with a recognition rate of 83.78%, which achieved a satisfactory recognition effect. Key words: cow posture recognition;image propcessing;feature extraction;Support Vector Machine;Neural Network 目录 摘要 1 关键词 1 Abstract 1 Key words 1 第1章 绪论 2 1.1 研究目的及意义 2 1.2 国内外研究现状 2 1.2.1人体姿态识别研究现状 2 1.2.2猪只姿态识别研究现状 3 1.3 论文内容及章节安排 3 1.3.1论文研究内容及技术路线 3 1.3.2论文的章节安排 4 1.4 本章小结 4 第2章 奶牛目标提取 5 2.1图像采集和图库建立 5 2.2奶牛目标图像分割 5 2.2.1基于传统图像分割算法的奶牛目标提取 5 2.2.2基于分水岭图像分割算法的奶牛目标提取 8 2.3.3基于支持向量机的奶牛目标提取 9 2.3本章小结 11 第3章 描述奶牛姿态的特征提取 12 3.1基于奶牛目标外轮廓的矩特征提取 12 3.1.1奶牛目标外轮廓的提取 12 3.1.2基于奶牛目标外轮廓的矩特征提取 13 3.2基于奶牛目标的几何特征提取 16 3.3本章小结 17 第4章 奶牛姿态识别 18 4.1基于支持向量机的奶牛姿态识别 18 4.1.1基于支持向量机的奶牛姿态识别步骤 18 4.1.2基于支持向量机的奶牛姿态识别结果与分析 19 4.2基于神经网络的奶牛姿态识别 19 4.2.1基于有监督学习神经网络的奶牛姿态识别 19 4.2.2基于无监督学习神经网络的奶牛姿态识别 21 (责任编辑：qin) |