Few-shot target correction technique base on Siamese network

• Main Authors: Zhang, Tao, Chen, JinLong, Yang, MingHao, Zhang, JiaQing, Zhai, WenZhe
• Format: Conference Proceeding
• Language: English
• Published: SPIE 12.05.2022
• DOI: 10.1117/12.2634513
• ISBN: 9781510653238; 1510653236
• ISSN: 0277-786X

In the process of Few-Shot detection, the traditional data enhancement method of Few-Shot expansion is used in most cases. However, in the process of sample enhancement, some object images will be generated which are quite different from the real object. This brings about the error detection caused by expanding the sample. Therefore, this paper proposes a target correction technology based on Siamese Network to offset this kind of situation. Based on the Siamese Network framework, this method adds a second decision module to the output end of the target detection algorithm to adjust the detection result and improve the detection effect under the condition of appropriately increasing the detection time cost. In particular, the error target and Few-Shot target in the detection results have a large deviation problem, and further improve the accuracy of the target detection algorithm under Few-Shot. The test results on UCAS-AOD and other public data sets show that the algorithm improves the accuracy of target detection algorithm for small samples.