Dangerous Target Recognition of Massive Image and Video Based on Deep Learning

The vast amount of data in the video requires people to check and identify, that is, it is necessary to extract a short possible image segment from a long-time video or massive image data. Technicians are prone to fatigue and long-term high-load work, and there are inevitable omissions, and computers can continue to work reliably as long as they have electricity. The paper is aimed at the dangerous targets that may appear in massive image and video data, especially unidentified personnel, vehicles, aircraft, etc. The computer image semantic recognition system automatically and quickly detect these potential dangers (or fragments of interest), and converts these massive “suspected” image languages into natural language that users can understand, and achieves early warning. The main work carried out in this paper is: label a certain amount of data with a dangerous target; on the platform of YOLOv3, a certain amount of training is carried out; we use the trained data for the identification of dangerous item, such as guns and knives, etc. The experiments shown that the dangerous image target is effectively identified.