农业工程学报
農業工程學報
농업공정학보
2015年
2期
14-21
,共8页
罗陆锋%邹湘军%熊俊涛%张宇%彭红星%林桂潮
囉陸鋒%鄒湘軍%熊俊濤%張宇%彭紅星%林桂潮
라륙봉%추상군%웅준도%장우%팽홍성%림계조
机器人%图像分割%定位%葡萄%采摘点
機器人%圖像分割%定位%葡萄%採摘點
궤기인%도상분할%정위%포도%채적점
robots%image segmentation%location%grape%picking point
针对葡萄果梗颜色复杂多变、轮廓不规则等影响因素使得采摘机器人难以准确对采摘点进行定位的问题，该文提出一种基于改进聚类图像分割和点线最小距离约束的采摘点定位新方法。首先通过分析葡萄图像的颜色空间，提取最能突显夏黑葡萄的HSI色彩空间分量H，运用改进的人工蜂群优化模糊聚类方法对葡萄果图像进行分割；然后对分割图像进行形态学去噪处理，提取最大连通区域，计算该区域质心、轮廓极值点、外接矩形；再根据质心坐标与葡萄簇边缘信息确定采摘点的感兴趣区域，在区域内进行累计概率霍夫直线检测，求解所有检测得出的直线到质心之间的距离，最后选取点线距离最小的直线作为采摘点所在线，并取线段中点坐标作为采摘点。以从晴天顺光、晴天遮阴、阴天光照下采集的300幅夏黑葡萄进行分类试验，结果表明，该方法的采摘点定位准确率达88.33%，平均定位时间为0.3467 s，可满足采摘机器人对采摘点的定位需求，为葡萄采摘机器人提供了一种新的采摘点求解方法。
針對葡萄果梗顏色複雜多變、輪廓不規則等影響因素使得採摘機器人難以準確對採摘點進行定位的問題，該文提齣一種基于改進聚類圖像分割和點線最小距離約束的採摘點定位新方法。首先通過分析葡萄圖像的顏色空間，提取最能突顯夏黑葡萄的HSI色綵空間分量H，運用改進的人工蜂群優化模糊聚類方法對葡萄果圖像進行分割；然後對分割圖像進行形態學去譟處理，提取最大連通區域，計算該區域質心、輪廓極值點、外接矩形；再根據質心坐標與葡萄簇邊緣信息確定採摘點的感興趣區域，在區域內進行纍計概率霍伕直線檢測，求解所有檢測得齣的直線到質心之間的距離，最後選取點線距離最小的直線作為採摘點所在線，併取線段中點坐標作為採摘點。以從晴天順光、晴天遮陰、陰天光照下採集的300幅夏黑葡萄進行分類試驗，結果錶明，該方法的採摘點定位準確率達88.33%，平均定位時間為0.3467 s，可滿足採摘機器人對採摘點的定位需求，為葡萄採摘機器人提供瞭一種新的採摘點求解方法。
침대포도과경안색복잡다변、륜곽불규칙등영향인소사득채적궤기인난이준학대채적점진행정위적문제，해문제출일충기우개진취류도상분할화점선최소거리약속적채적점정위신방법。수선통과분석포도도상적안색공간，제취최능돌현하흑포도적HSI색채공간분량H，운용개진적인공봉군우화모호취류방법대포도과도상진행분할；연후대분할도상진행형태학거조처리，제취최대련통구역，계산해구역질심、륜곽겁치점、외접구형；재근거질심좌표여포도족변연신식학정채적점적감흥취구역，재구역내진행루계개솔곽부직선검측，구해소유검측득출적직선도질심지간적거리，최후선취점선거리최소적직선작위채적점소재선，병취선단중점좌표작위채적점。이종청천순광、청천차음、음천광조하채집적300폭하흑포도진행분류시험，결과표명，해방법적채적점정위준학솔체88.33%，평균정위시간위0.3467 s，가만족채적궤기인대채적점적정위수구，위포도채적궤기인제공료일충신적채적점구해방법。
In wine brewing process, the most time-consuming and laborious is grape picking. In order to improve the target location accuracy and work efficiency of grape-picking robot, reduce the mechanical damage that was caused by improper positioning of grape picking point, in view of some influence factors such as the various colors of grape stem and the irregular contour of grape, which make picking robot hard to locate picking point accurately, a new method of picking point location based on the improved image segmentation with clustering and the constraint by minimum distance between dot and line was put forward in this paper. Because the picking time was often chosen in sunny or cloudy day, 300 images of summer black grape were collected using D5200 Nikon digital camera in sunny or cloudy days, which were taken as test materials; the shooting distance between camera and grape cluster was about 80 cm, and the sizes of these images were zoomed to 800×600 pixels. Firstly, the color space of gathered images was analyzed and the component H of HSI color space that can mostly highlight summer black grape was found. The H component of images was extracted and median filtering was performed on these images. Grape fruit image was segmented by using fuzzy clustering that was improved by artificial swarm. Solving the minimum value of fuzzy objective function of FCM clustering algorithm was transformed into solving the maximum value of artificial swarm fitness function by improving the fitness function of the artificial swarm optimization algorithm. Then, the segmented image was processed with morphological denoising, the maximum connected region was extracted, and the regional barycenter, extreme point and external rectangular were calculated. Secondly, the interest region of picking point was determined according to the barycentric coordinates and edge information of grape image. Taking 0.6 times of the length of the external rectangular as the length of region of interest (ROI), and taking 0.5 times of the vertical distance between hightest point of profile and barycenter as the height of ROI, the image of ROI was preprocessed with bilateral filter, and its edge image was extracted through gradient edge detection to obtain binary edge. And Hoffe’s straight line testing of cumulative probability was taken in this ROI, maximum votes were taken to 18, all the detected endpoints of line were recorded, and linear equation was established by the two endpoints; the distances between all the detected straight lines and barycenter were solved by using the method of shortest distance from point to straight line, the straight line whose distance to barycenter was the smallest was chosen as the line where picking point was located, and the midpoint’s coordinates of elected line was used as picking point. Finally, in order to verify the method proposed in this paper, a classification experiment based on 300 images of summer black grape gathered under direct sunlight, backlighting in sunny day and overcast light was carried out, using Opencv2.3.1 and Visual C++ as the algorithm programming platform. Through the statistical analysis of 300 testing images, it was discovered that the ratio of location error within 12 pixels between obtained picking point using this paper’s method and optimal picking point by hand could reach up to 88.33%, and these obtained points were located in the stems of grapes, which could meet the requirements of picking robot’s location. When grape clusters were located in both sides of image frame and their covered parts were less than one-third, and meanwhile the stem located in the ROI had not been covered mostly, the method could be well to solve the position of grape picking point. The average elapsed time of picking point’s positioning in natural conditions was 0.3467 seconds. The test results of location error and elapsed time show that the method proposed in this paper can meet the demand of picking robot on picking point’s positioning and is a new solution method of picking point for grape-picking robot.
