Immune coordination deep network for hand heat trace extraction

Abstract

The recognition and detection of infrared heat trace plays an important role in the search and tracking fields of criminal investigation and military. However, due to the irregular contact area and fuzzy depth of heat trace, the existing methods cannot accurately extract the trace target. Based on the biological immune coordination mechanism, an immune coordination deep network (ICDNet) for hand heat trace extraction is proposed in this paper. It is composed of small-scale intelligent recognition method based on immune system coordination mechanism and neural immune coordination recognition method based on neural system and immune system coordination mechanism. The former divides the fuzzy area of the heat trace into multiple image patches to realize feature extraction and classification of corresponding pixels. The latter extracts the characteristics of heat trace, discriminates the type of heat trace, and obtains the prior information of heat trace. Combined with the recognition results, the infrared heat trace target is extracted. Extensive experiments on the infrared hand trace dataset demonstrate the effectiveness of our method. We achieve 91.50% mIoU on the test set, which is 23.65% higher than the latest methods.