President Chain Store Shelves Commodity Segmentation Study And Identification

In this modern society, there is a wide variety of commodities. Therefore, keeping track of the sale and inventory status of the commodities is essential to understanding the operation of the entire market.The traditional supermarket managing software mostly used for query and modification of categories and numbers of static goods, which requires substantial human input ahead and timely supplementary information update afterwards. Thus, these softwares have the disadvantages such as lagging information update and large demand for labor force. Further study on methods of acquiring and analyzing supermarket commodities data needed.Aiming to solve the problems mentioned above, this article utilizes knowledge on digital picture processing technique and machine learning, and comes up with a new and easily available identification method of acquiring basic information of the commodities on the shelf.Firstly, a threshold was identified according to the data of several experiments. Then, level hierarchical segmentation processing was done successfully combining with shelves binary image horizontal projection histogram in order to breakup the whole imaging processing into parts. Then integration analysis was done on horizontal projection of different color space components, and columns of the image segmentation of each layer were extracted. Afterwards, commodity target recognition areas were repeatedly corrected using the column line and the image characteristics of the sample library. Finally, color, shape and location characteristics of the commodities were extracted, target recognition areas were recognized and judged using SVM classifier, and monomer segmentation and image information summarized statistics were done.The study tested the shelf of the beverage area, differentially selected various image sample sets, several standard test images, and compared and analyzed recognition time and recognition rates under different experiment conditions. The result indicated that recognition time varied among different test images, and different color characteristics of the same test image had large influence on the recognition time and recognition rates. However, above a certain size, further changes of the size of sample library had few influence on the recognition results.