Automatic annotation algorithm based on sliding window moment feature matching

Abstract

In the era of big data, detecting and identifying targets based on massive data sources is a very important task. At present, most of the labeling of large-scale image data relies on traditional manual labeling methods, which takes a long time and is inefficient. In order to efficiently construct large-scale maritime target image data sets, we propose an automatic annotation algorithms, namely: automatic annotation algorithm based on moment features. Experiments were conducted to verify the accuracy of the automatic annotation algorithm to annotate image data, and finally proved that the two image automatic annotation algorithms proposed by us can construct the marine target image data set more efficiently, and provide good data support for downstream tasks.