Detection of small targets in aerial images is still a difficult problem due to the low resolution and background-like targets. With the recent development of object detection technology, efficient and high-performance detector techniques have been developed. Among them, the YOLO series is a representative method of object detection that is light and has good performance. In this paper, we propose a method to improve the performance of small target detection in aerial images by modifying YOLOv5. The backbone is was modified by applying the first efficient channel attention module, and the channel attention pyramid method was proposed. We propose an efficient channel attention pyramid YOLO (ECAP-YOLO). Second, in order to optimize the detection of small objects, we eliminated the module for detecting large objects and added a detect layer to find smaller objects, reducing the computing power used for detecting small targets and improving the detection rate. Finally, we use transposed convolution instead of upsampling. Comparing the method proposed in this paper to the original YOLOv5, the performance improvement for the mAP was 6.9% when using the VEDAI dataset, 5.4% when detecting small cars in the xView dataset, 2.7% when detecting small vehicle and small ship classes from the DOTA dataset, and approximately 2.4% when finding small cars in the Arirang dataset.

This paper presents a region-adaptive clutter rejection method for small target detection in sea-based infrared search and track. In the real world, clutter normally generates many false detections that impede the deployment of such detection systems. Incoming targets (missiles, boats, etc.) can be located in the sky, horizon and sea regions, which have different types of clutters, such as clouds, a horizontal line and sea-glint. The characteristics of regional clutter were analyzed after the geometrical analysis-based region segmentation. The false detections caused by cloud clutter were removed by the spatial attribute-based classification. Those by the horizontal line were removed using the heterogeneous background removal filter. False alarms by sun-glint were rejected using the temporal consistency filter, which is the most difficult part. The experimental results of the various cluttered background sequences show that the proposed region adaptive clutter rejection method produces fewer false alarms than that of the mean subtraction filter (MSF) with an acceptable degradation detection rate.

A novel temporal contrast filter (TCF)‐based method was developed to detect supersonic small infrared targets. The proposed hysteresis threshold‐based detection followed by the TCF can enhance the accuracy of the target position, the robustness to background clutter and the velocity range of moving targets compared with the conventional temporal variance filter‐based method.

Long-range ground targets are difficult to detect in a noisy cluttered environment using either synthetic aperture radar (SAR) images or infrared (IR) images. SAR-based detectors can provide a high detection rate with a high false alarm rate to background scatter noise. IR-based approaches can detect hot targets but are affected strongly by the weather conditions. This paper proposes a novel target detection method by decision-level SAR and IR fusion using an Adaboost-based machine learning scheme to achieve a high detection rate and low false alarm rate. The proposed method consists of individual detection, registration, and fusion architecture. This paper presents a single framework of a SAR and IR target detection method using modified Boolean map visual theory (modBMVT) and feature-selection based fusion. Previous methods applied different algorithms to detect SAR and IR targets because of the different physical image characteristics. One method that is optimized for IR target detection produces unsuccessful results in SAR target detection. This study examined the image characteristics and proposed a unified SAR and IR target detection method by inserting a median local average filter (MLAF, pre-filter) and an asymmetric morphological closing filter (AMCF, post-filter) into the BMVT. The original BMVT was optimized to detect small infrared targets. The proposed modBMVT can remove the thermal and scatter noise by the MLAF and detect extended targets by attaching the AMCF after the BMVT. Heterogeneous SAR and IR images were registered automatically using the proposed RANdom SAmple Region Consensus (RANSARC)-based homography optimization after a brute-force correspondence search using the detected target centers and regions. The final targets were detected by feature-selection based sensor fusion using Adaboost. The proposed method showed good SAR and IR target detection performance through feature selection-based decision fusion on a synthetic database generated by OKTAL-SE.

Deep learning technology for real-time small object detection in aerial images can be used in various industrial environments such as real-time traffic surveillance and military reconnaissance. However, detecting small objects with few pixels and low resolution remains a challenging problem that requires performance improvement. To improve the performance of small object detection, we propose DCEF${ }^2$-YOLO. Our proposed method enables efficient real-time small object detection by using a deformable convolution (DFConv) module and an efficient feature fusion structure to maximize the use of the internal feature information of objects. DFConv preserves small object information by preventing the mixing of object information with the background. The optimized feature fusion structure produces high-quality feature maps for efficient real-time small object detection while maximizing the use of limited information. Additionally, modifying the input data processing stage and reducing the detection layer to suit small object detection also contributes to performance improvement. When compared to the performance of the latest YOLO-based models (such as DCN-YOLO and YOLOv7), DCEF${ }^2$-YOLO outperforms them, with a mAP of +6.1% on the DOTA-v1.0 test set, +0.3% on the NWPU VHR-10 test set, and +1.5% on the VEDAI512 test set. Furthermore, it has a fast processing speed of 120.48 FPS with an RTX3090 for 512 × 512 images, making it suitable for real-time small object detection tasks.