Objective The polarization information is related to the material properties and physicochemical characteristics of objects, with different objects exhibiting distinct polarization characteristics. This unique property endows infrared polarization technology with significant advantages in the recognition of camouflaged targets. To meet the application requirements of recognition technology for high-temperature burning camouflage in the air, this paper conducts a study on the long-wave infrared polarization recognition technology for high-temperature solid/gas targets based on biomimetic polarization vision principles.

Methods Taking high-temperature metal heat sources and butane gas combustion heat sources as research subjects, the two-channel PD2 (Fig.1), three-channel PD3 (Fig.2), and four-channel PD4 polarization distance models (Fig.3) are used for imaging high-temperature solid/gas heat source scenarios (Fig.6). The polarization characteristics and target-background contrast of high-temperature solid heat sources are analyzed using four quality evaluation metrics. The three channel polarization distance images are encoded and fused using R, G, and B channels, combining the polarization information of each channel. Three channel combination forms, 234, 223, and 443, are selected for target-background contrast enhancement, and the target-background contrast is analyzed using quality evaluation metrics. A dataset of three types of fused images is created, and the YOLOv8 network is used for solid/gas heat source target recognition.

Results and Discussions For PD2 (Fig.6(b)), the camouflaged heat source exhibits edge polarization information. For PD3 (Fig.6(c)) and PD4 (Fig.6(d)), the edge and texture information of the camouflaged heat source are displayed, and there is a significant difference between the target and background in the images. According to the quality evaluation function (Tab.1), PD2 has a BSF value of 1.8236, indicating strong background suppression capability. PD3 and PD4 have high GSC, SCR, and FD indicator values, showing good contrast in noisy environments, with PD3 performing the best. PD3, which has excellent target-background contrast, is introduced into PD2 and PD4 for contrast enhancement. According to the quality evaluation function (Tab.2), the 223 combination (Fig.7(b))'s BSF value is reduced to 0.6238 compared to PD2, making the solid/gas heat source target more prominent. Compared to PD4, the 443 combination (Fig.7(c))'s GSC, SCR, and FD values increased, enhancing the target-background contrast. The 234 combination (Fig.7(a)) has the highest GSC, SCR, and FD values among the three fused images, performing the best. Solid/gas heat source targets can be effectively identified in the dataset of the three types of fused polarization distance infrared images (Fig.8). According to Tab.3, their mAP_0.5 is over 99%, with mAP_0.5∶0.95 being 73.7%, 75.2%, and 67.3%, respectively. The 223 combination dataset, due to the introduction of PD3, has a reduced BSF value and less background clutter, resulting in the highest mAP_0.5∶0.95. The 234 dataset, which integrates PD2, PD3, and PD4, has rich polarization information and less background clutter, with a detection accuracy of 73.7%. The 443 combination has the lowest mAP_0.5∶0.95 value due to more background clutter.

Conclusions This paper conducts a study on the polarization recognition technology for high-temperature solid/gas targets in the long-wave infrared band. By combining bionic vision algorithms, the extraction of material and edge information of high-temperature solid/gas targets can be effectively enhanced, increasing the accuracy of classification and recognition. The integration of bionic polarization vision algorithms with deep learning technology enables a comparative study of classification and recognition of high-temperature solid/gas targets under multi-channel combinations.