基于最小二乘支持向量机的梅花鹿进食行为识别的研究

发布时间：2018-05-30 20:38

本文选题：最小二乘支持向量机 + 机器学习　；参考：《吉林农业大学》2017年硕士论文

【摘要】：我国养鹿业历史悠久,但在国际市场中所占份额远低于西方国家。对梅花鹿生态行为进行研究,有助于进一步开发鹿产品的经济价值,更好的对梅花鹿进行保护,为智能农业提供技术支撑。图像处理识别技术在科学养殖方面也有着广泛应用,但其中多数集中于禽类等小型养殖领域,应用于禽畜养殖研究较少,对于牛、羊、鹿等大型畜牧类动物的研究基本没有。本文拟用计算机图像识别技术运用到梅花鹿的科学养殖中,节约了人工养殖成本。论文还对图像进行各种细小环节的处理。如二值化,归一化,细化等等,提高了图像处理的质量,以便机器进行分析识别。论文研究的具体内容如下:(1)对最小二乘支持向量机的原理及其基本算法进行探究,并根据应用需要,选择SVM中的经典多分类算法OVA(One-Versus-All)解决梅花鹿视频抓取图像与数据库图像的渐进性匹配问题,因为在相同情况下,识别精度较高。(2)解决过程中,着重分析了特征值提取办法及参数设置方法,为提高系统识别准确率,提出了一种适用于小样本特征提取的解决算法SLDA,经过实验验证,可以有效提高特征提取的准确率,为后续的研究奠定了坚实的基础。并将研究结果有效应用于梅花鹿面部图像识别这一应用领域中。(3)为了增强整个系统识别的精度,对图像进行预先处理,去除一些对图像识别可能存在的干扰的情况。(4)根据系统总体设计方案构建梅花鹿远程视频解析及识别系统平台,系统主要功能分别是,预览功能,实时监测功能,面部检测功能,面部识别功能及面部训练功能等及部分。并将其应用在梅花鹿科学养殖中。
[Abstract]:China has a long history of deer farming, but its share in the international market is much lower than that in western countries. The study on the ecological behavior of sika deer is helpful to further develop the economic value of deer products, better protect sika deer, and provide technical support for intelligent agriculture. Image processing and recognition techniques are also widely used in scientific breeding, but most of them are concentrated in the field of small-scale breeding such as poultry, but few are used in livestock breeding, but there is no research on large livestock animals such as cattle, sheep, deer and so on. In this paper, the technology of computer image recognition is applied to the scientific breeding of sika deer, which saves the cost of artificial breeding. The paper also carries on the processing to the image each kind of small link. Such as binarization, normalization, thinning and so on, improve the quality of image processing, so that the machine can be analyzed and identified. The specific contents of this paper are as follows: (1) the principle and basic algorithm of least squares support vector machine (LS-SVM) are explored, and the application needs are also discussed. The classical multi-classification algorithm OVAX One-Versus-Allin SVM is chosen to solve the problem of progressive matching between sika deer video capture image and database image, because in the same case, the recognition accuracy is higher. The methods of feature extraction and parameter setting are analyzed emphatically. In order to improve the recognition accuracy of the system, a new algorithm, SLDA, which is suitable for small sample feature extraction, is proposed, which can effectively improve the accuracy of feature extraction after experimental verification. For the follow-up research laid a solid foundation. In order to enhance the recognition accuracy of the whole system, the research results are effectively applied to the face image recognition of sika deer. Removing some possible interference to image recognition.) according to the overall design of the system, the platform for remote video analysis and recognition of sika deer is constructed. The main functions of the system are: preview function, real-time monitoring function, Facial detection function, facial recognition function and facial training function and so on. It was applied to sika deer breeding.
【学位授予单位】：吉林农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S825;TP391.41

【参考文献】