Small Low-Contrast Target Detection Data-Driven Spatiotemporal Feature Fusion and Implementation
Abstract : Detecting small low-contrast targets in the airspace is an essential and challenging task. This article proposes a simple and effective data-driven support vector machine (SVM)-based spatiotemporal feature fusion detection method for small low-contrast targets. We design a novel pixel-level feature, called a spatiotemporal profile, to depict the discontinuity of each pixel in the spatial and temporal domains The spatiotemporal profile is a local patch of the spatiotemporal feature maps concatenated by the spatial feature maps and temporal feature maps in channelwise, which are generated by the morphological black-hat filter and a ghost-free dark-focusing frame difference methods, respectively. Instead of the handcrafted feature fusion mechanisms in previous works, we use the labeled spatiotemporal profiles to train an SVM classifier to learn the spatiotemporal feature fusion mechanism automatically. To speed up detection for high-resolution videos, the serial SVM classification process on central processing units (CPUs) is reformed as parallel convolution operations on graphics processing unit (GPUs), which exhibits over 1000+ times speedup in our real experiments. Finally, blob analysis is applied to generate final detection results. Elaborate experiments are conducted, and experimental results demonstrate that the proposed method performs better than 12 baseline methods for the small low-contrast target detection. The field tests manifest that the parallel implementation of the proposed method can realize real-time detection at 15.3 FPS for videos at a resolution of 2048x1536 and the maximum detection distance can reach 1 km for drones in sunny weather.
? Existing open-source dataset in infrared small target detection is scarce, most traditional methods are evaluated on their in-house datasets.
? However, existing CNNbased methods cannot be directly applied for infrared small targets since pooling layers in their networks could lead to the loss of targets in deep layers.
? Compared to existing methods, our method is more robust to the variations of clutter background, target size, and target shape.
? Existing CNN-based methods can be divided into detection based methods and segmentation based methods.
? Existing CNNbased methods mainly use real infrared data with manual annotations.
? Detecting infrared small targets is thus a form of blob detection, which is a problem with a long history in the image processing literature.
? However, in real-world scenarios, the problem is more complex: there are more distractors that also stand out as outliers in the background.
? Infrared small target detection can also be modeled as a supervised machine-learning problem, but it has long been stuck with insufficient training data due to the difficulty of collecting infrared small target images.
? We follow the convention of this field and model infrared small target detection as a segmentation problem.
• This paper proposed a hybrid method by making a compromise for the spatial filter approach and temporal approach, known as the separate spatio-temporal filtering method based on an attribute-based plot association.
• In contrast, the proposed method (horizontal clutter rejection (L-DBRF) after M-MSF) showed an ideal ROC curve pattern.
• To detect infrared small targets, numerous traditional methods have been proposed, including filtering-based methods, local-contrast-based methods, and low-rank based methods.
• A dense nested interactive module and a channel-spatial attention module are proposed to achieve progressive feature fusion and adaptive feature enhancement.
? We also compare the performance of our network against other model-driven methods and deep networks on the open SIRST dataset as well.
? It can be seen that, compared with FPN, the performance of the DLC-FPN is consistently and significantly better.
? It can be seen that by covering multiple dilation rates, the performance of MLCFPN is consistently better than DLC-FPN.
? The comparison among FPN, DLC-FPN, and MLC-FPN suggests that incorporating local contrast prior helps and a multi-scale measure in the same layer can further boost the performance.
? All convolutional networks perform better than the non-learning model-driven methods, which shows that learning from the data offers a promising way leading to better performance for infrared small detection.
|