Research On Ripe Tomato Vision Technique Of Harvesting Robot

Recently, the development of agriculture is heading to mass production, diversification and precision. The cost of labor force in the field is getting higher, and the phenomenon of lack of labor force is getting serious. Therefore, more and more people pay attention to the research on agriculture robot. In the research of tomato harvesting robot, there are two very important issues: the instability of image quality brought about illumination and robot real-time. In the paper, the following research achievements have been fulfilled:(1) Experiment Design. Many methods of image deepness, the method of two lenses measuring distance is suitable for our experiment system. The hardware and software of robot vision system is introduced. And the environment and precept of the experiment is designed.(2) Image Pre-processing. The light compensating method and the HSItogram equalization algorithms are introduced. Moreover, the Prewitt algorithm is selected based on the color image processing results of the several image-sharpening algorithms. Several image smoothing algorithms, including Median-Filter K-Nearest-Neighbor Filter and Symmetric Nearest Neighbor Filter, are brought forward to clear up the noises in the image, and compared with their experiment results, it is proved that the Median-Filter algorithm is effective because of it's real-time and easy calculation, In our experiment system, therefore, the Median-Filter algorithm is adopted.(3) Image Segment and Recognition. Compare with all kinds of color- spaces, the HSI color- space and the YUV color- space, which can separate the intensity from color, are selected based on the feature of robot vision. As a result, the Hue-HSItogram Statistic Double-threshold Algorithm Based on HSI Color-space and the Double-threshold Algorithm Based on YUV Color- space are put forward. And it is proved by experiment that the latter algorithm has the better real-time and the better effect than the other.(4) Image Sorting and Feature Extracting. The Dilation and the Erosion algorithm, which are Binary morphology algorithm, and the Seed-Filling algorithm, are selected to clear up the noises in the segmented image. In the feature extracting section, the two features, object area and object center, are abstracted. It is prepared for the following project, which will position the object, to extract the object center feature.(5) Based on above algorithms, the harvesting robot ripe tomato vision recognition software (RobotVision 1.0) is designed to validate those algorithms.In this paper, a new approach for ripe tomato recognition is pointed out. It is of significance for developing agriculture robot research and raising agriculture international competition in our country.