Research On The Algorithm Of Counting Adhesive Objects

Along with the rapid development of the computer technology, the artificial intelligence and the mind science research, digital image processing goes for the higher and deeper level.It was just used for improving the image quality in the early time, however, with the time going, it takes the problem of how to explain images by computer system and how to realize the similar humanity vision system to understand the outside world the stage as the main goal.In the sericulture domain, the counting of the silkworm ovum automatically has been a big difficulty which restricts its development. At present, We hit the target of counting the silkworm ovum's unit weights by the naked eye discernment manually. But for the silkworm ovum is small and its quantity are so many that it leads to the big work load and tasteless monotonous feeling for the operators. Moreover, if operated improperly, it is also easy to damage the silkworm ovum which immediately influences the economic efficiency. But if takes the computer as the carrier exploring the digital image processing method, it could improve the counting efficiency and reduces operators' labor intensity. At the same time, the testing result is more objective and the detecting process automation is enhanced with the auxiliary function of computer's data management simultaneously. Therefore, for the automatic, objective and the fast characteristic of the digital image processing technology in the counting of silkworm ovum automatically, it becomes research key in the world.The main target of the digital image processing is an image, but for the reason of noise, illumination as well as object's mutual adhesion situations and so on, the quality of the gathering images will always be influenced in different degree, therefore, how to figure out this kind of images is key study point in this article. The traditional process method is to use the image segmentation technology firstly, and then carries on the following processing. But if it has the adhesion or the part overlap phenomenon in the image, it often causes the incomplete division. So it brings big trouble for the following image analysis processing. To solve this problem, a new algorithm based on mesh model is proposed. It makes full use of machine-learning to avoid the process of segmentation, and then the counting result is given directly by SVM. Moreover, it also gives an extraction of effective texture feature and improves the precision of SVM. Compared with the traditional algorithm based on segmentation, experimental results show the high performance in both robustness and precision of the proposed method.