Recent deep models trained on large-scale RGB datasets lead to considerable achievements in visual detection tasks. However, the training examples are often limited for an infrared detection task, which may deteriorate the performance of deep detectors. In this paper, we propose a transfer approach, Source Model Guidance (SMG), where we leverage a high-capacity RGB detection model as the guidance to supervise the training process of an infrared detection network. In SMG, the foreground soft label generated from the RGB model is introduced as source knowledge to provide guidance for cross-domain transfer. Additionally, we design a Background Suppression Module in the infrared network to receive the knowledge and enhance the foreground features. SMG is easily plugged into any modern detection framework, and we show two explicit instantiations of it, SMG-C and SMG-Y, based on CenterNet and YOLOv3, respectively. Extensive experiments on different benchmarks show that both SMG-C and SMG-Y achieve remarkable performance even if the training set is scarce. Compared to advanced detectors on public FLIR, SMG-Y with 77.0% mAP outperforms others in accuracy, and SMG-C achieves real-time detection at a speed of 107 FPS. More importantly, SMG-Y trained on a quarter of the thermal dataset obtains 74.5% mAP, surpassing most state-of-the-art detectors with full FLIR as training data.

Robust infrared (IR) small target detection is critical for infrared search and track (IRST) systems and is a challenging task for complicated backgrounds. Current algorithms have poor performance on complex backgrounds, and there is a high false alarm rate or even missed detection. To address this problem, a weighted local coefficient of variation (WLCV) is proposed for IR small target detection. This method consists of three stages. First, the preprocessing stage can enhance the original IR image and extract potential targets. Second, the detection stage consists of a background suppression module (BSM) and a local coefficient of variation (LCV) module. BSM uses a special three-layer window that combines the anisotropy of the target and differences in the grayscale distribution. LCV exploits the discrete statistical properties of the target grayscale. The weighted advantages of the two modules complement each other and greatly improve the effect of small target enhancement and background suppression. Finally, the weighted saliency map is subjected to adaptive threshold segmentation to extract the true target for detection. The experimental results show that the proposed method is more robust to different target sizes and background types than other methods and has a higher detection accuracy.

Image super-resolution (SR) algorithms based on deep learning yield good visual performances on visible images. Due to the blurred edges and low contrast of infrared (IR) images, methods transferred directly from visible images to IR images have a poor performance and ignore the demands of downstream detection tasks. Therefore, an Inception Dilated Super-Resolution (IDSR) network with multiple branches is proposed. A dilated convolutional branch captures high-frequency information to reconstruct edge details, while a non-local operation branch captures long-range dependencies between any two positions to maintain the global structure. Furthermore, deformable convolution is utilized to fuse features extracted from different branches, enabling adaptation to targets of various shapes. To enhance the detection performance of low-resolution (LR) images, we crop the images into patches based on target labels before feeding them to the network. This allows the network to focus on learning the reconstruction of the target areas only, reducing the interference of background areas in the target areas’ reconstruction. Additionally, a feature-driven module is cascaded at the end of the IDSR network to guide the high-resolution (HR) image reconstruction with feature prior information from a detection backbone. This method has been tested on the FLIR Thermal Dataset and the M3FD Dataset and compared with five mainstream SR algorithms. The final results demonstrate that our method effectively maintains image texture details. More importantly, our method achieves 80.55% mAP, outperforming other methods on FLIR Dataset detection accuracy, and with 74.7% mAP outperforms other methods on M3FD Dataset detection accuracy.

In the field of image fusion, the integration of infrared and visible images aims to combine complementary features into a unified representation. However, not all regions within an image bear equal importance. Target objects, often pivotal in subsequent decision-making processes, warrant particular attention. Conventional deep-learning approaches for image fusion primarily focus on optimizing textural detail across the entire image at a pixel level, neglecting the pivotal role of target objects and their relevance to downstream visual tasks. In response to these limitations, TDDFusion, a Target-Driven Dual-Branch Fusion Network, has been introduced. It is explicitly designed to enhance the prominence of target objects within the fused image, thereby bridging the existing performance disparity between pixel-level fusion and downstream object detection tasks. The architecture consists of a parallel, dual-branch feature extraction network, incorporating a Global Semantic Transformer (GST) and a Local Texture Encoder (LTE). During the training phase, a dedicated object detection submodule is integrated to backpropagate semantic loss into the fusion network, enabling task-oriented optimization of the fusion process. A novel loss function is devised, leveraging target positional information to amplify visual contrast and detail specific to target objects. Extensive experimental evaluation on three public datasets demonstrates the model’s superiority in preserving global environmental information and local detail, outperforming state-of-the-art alternatives in balancing pixel intensity and maintaining the texture of target objects. Most importantly, it exhibits significant advantages in downstream object detection tasks.