基于纹理颜色的温室大棚黄瓜图像识别

发布时间：2018-06-04 17:48

本文选题：图像处理 + 最大稳定极值区域　；参考：《天津理工大学》2017年硕士论文

【摘要】：我国是农业大国,农业人口在世界上最多,农业是我国赖以生存的基础产业。我国的耕地面积仅占世界7%,却要承载占世界1/5的人口。随着我国老龄化趋势的加剧,导致从事农业生产的人口逐年下降,因此解决这一问题刻不容缓。随着机器人技术和人工智能的发展,机器人进入农业生产领域成为可能。引进农业机器人进行农业生产可以有效的缓解当前矛盾,在促进机器人行业发展的同时解放有限的劳动力资源,还可以大大降低黄瓜的生产成本,提高生产率。农业机器人的机械收获需要自动检测和计数树冠的果实。由于颜色相似,形状不规则,背景复杂等问题造成果实识别非常困难而难以执行采摘动作。因此基于所有上述问题,在复杂背景下识别黄瓜是一项具有挑战性的任务。在本文中,提出了一种基于纹理分析和颜色分析的技术用于检测温室中的黄瓜。RGB图像转换为灰度图像和HSI图像分别执行算法。在第一阶段进行颜色分析以除去背景,例如土壤,树枝和天空,同时保持尽可能多的绿色水果像素呈现黄瓜和叶子。同时,MSER和HOG应用于灰度图像中的纹理分析。通过MSER获得候选区域,这些候选区域有的是包括黄瓜的候选区域。HOG+SVM用于区分黄瓜目标和背景消除不是黄瓜的候选区域。为了进一步消除非目标区域,通过SIFT算法检测关键点,根据SIFT的疏密特征确定黄瓜区域。然后,将颜色分析和纹理分析的结果合并以获得候选黄瓜区域。最后阶段应用数学形态学操作以获得完整的黄瓜。经过以上操作后可以得到相较于一般分割方法更加精确的黄瓜轮廓,在农业大棚中的自然光照环境下进行测试识别率高于普通的分割方法,保证了农业机器人后续的准确采摘。
[Abstract]:China is a large agricultural country with the largest agricultural population in the world. Agriculture is the basic industry on which our country depends for survival. China's arable land accounts for only 7 percent of the world's population, but it is responsible for a fifth of the world's population. With the aggravation of the aging trend in China, the population engaged in agricultural production is decreasing year by year, so it is urgent to solve this problem. With the development of robot technology and artificial intelligence, it is possible for robots to enter the field of agricultural production. The introduction of agricultural robots to agricultural production can effectively alleviate the current contradictions, promote the development of the robot industry while liberating limited labor resources, but also greatly reduce the production costs of cucumber, improve productivity. The mechanical harvesting of agricultural robots requires automatic detection and counting of tree crown fruits. Because of similar color, irregular shape and complex background, fruit recognition is very difficult to carry out picking. Therefore, it is a challenging task to identify cucumber in complex background based on all the above problems. In this paper, a technique based on texture analysis and color analysis is proposed to detect the conversion of cucumber image into gray-scale image and HSI image in greenhouse. Color analysis is performed in the first phase to remove the background, such as soil, branches and sky, while maintaining as many green fruit pixels as possible to present cucumber and leaves. At the same time, MSER and HOG are applied to texture analysis in grayscale images. Candidate regions were obtained by MSER. These candidate regions include cucumber candidate regions. Hog SVM is used to distinguish between cucumber target and background elimination is not a cucumber candidate region. In order to eliminate the non-target region further, the key points were detected by SIFT algorithm, and the cucumber area was determined according to the density characteristics of SIFT. The results of color analysis and texture analysis were combined to obtain candidate cucumber regions. In the final stage, mathematical morphology was used to obtain the complete cucumber. After the above operation, the cucumber contour is more accurate than the general segmentation method, and the test recognition rate in the natural light environment in the agricultural greenhouse is higher than that in the ordinary segmentation method, which ensures the accurate picking of the agricultural robot.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S626;S642.2;TP391.41

【相似文献】