Unmanned Aerial Vehicles (UAVs) have emerged as versatile tools across various industries, providing valuable insights through aerial image analysis. However, the efficacy of UAV-deployed image detection systems is often limited by the resolution of captured images and the altitudinal constraints of UAV operations. This paper introduces a novel integration of the detection system with super-resolution networks and image reconstruction techniques, inspired by the exceptional visual capabilities of eagles, to enhance image detail and detection recall from aerial perspectives. The super-resolution component utilizes advanced algorithms to upscale the resolution of images captured by UAVs, thereby improving the granularity and clarity of the visual data. Concurrently, image reconstruction techniques are applied to enhance the quality of original images further. Additionally, we propose an innovative adaptive feature fusion (AFF) technique, which not only surpasses traditional concatenation methods in integrating multiscale features effectively but also demonstrates remarkable improvement in feature utilization and further refinement of the fusion process. Extensive experiments conducted on VisDrone2019 and DOTA datasets demonstrate that our integrated system significantly outperforms existing methods in terms of detection precision and recall. Compared to YOLOv5s, Recall and mAP50 have increased by 8.89% and 11.11% respectively, with only a slight increase in the number of parameters and complexity

Object detection in unmanned aerial vehicle (UAV) imagery has been widely applied across various domains. However, the typically small size and uneven spatial distribution of objects in UAV imagery pose significant challenges for UAV object detection tasks. Confronting such challenges, we propose Eagle-YOLOv8, an object detection algorithm for UAV imagery inspired by the eagle-eye vision system. Firstly, inspired by the double fovea mechanism of the eagle eye, we construct a long-focus attention (LFA) module, which can promote the network to focus on the target and pay more attention to discriminative features. Secondly, we propose a feature weight fusion network (FWFN) inspired by the double field of view characteristics of eagle eye. This network utilizes a novel weight fusion technique to alternative the conventional concatenate method, which assigns weights to feature layers according to their importance. Finally, we analyze the effect of Wise-IoU loss on the fit of the prediction box to the object. In addition, we create a dataset called AerialDet with eight categories to validate the generalization performance of the proposed method. Experimental evaluations conducted on both the challenging VisDrone2019-Det dataset and our self-collected dataset validate the effectiveness of Eagle-YOLOv8. The proposed method outperforms baseline approaches in object detection performance, exhibiting notable improvements: 8.56% in Precision metrics, 10.06% in mAP50 metrics, and 9.43% in Recall metrics, all achieved with only a marginal increase in parameters compared to YOLOv8 (small). Our code is available at: https://github.com/772736243/zehaoGao

This paper addresses the challenge of 3D object detection from a single panoramic image under severe deformation. The advent of the two-stage approach has impelled significant progress in 3D object detection. However, most available methods only can localize region proposals by a single-scale architecture network, which are sensitive to deformation and distortion. To address this issue, we propose a multi-scale convolutional neural network (MSCNN) to estimate the 3D pose of an object. To be specific, the proposed MSCNN consists of three steps for effectively detecting the distorted object on the panoramic images. The MSCNN contains the CycleGAN network that converts rectilinear images into panoramas, a fused framework that improves both accuracy and speed for object detection, and an adversarial spatial transformer network (ASTN) that extracts the deformation features of the object on panoramic images. Additionally, we recover the 3D pose of the object using a coordinate projection and a 3D bounding box. Extensive experiments demonstrate that the proposed method can achieve a 3D detection accuracy of 38.7% in high-resolution panoramic images, which is higher than the current state-of-the-art algorithm of 5.2%. Moreover, the speed of detection is only about 0.6 seconds per image, which is six times faster than Faster R-CNN (COCO). The code will be available at https://github.com/Yanhui-He.

As a fundamental preprocessing technique, remote sensing image motion deblurring is important for visual understanding tasks. Most conventional approaches formulate the image motion deblurring task as a kernel estimation. Because the kernel estimation is a highly ill-posed problem, many priors have been applied to model the images and kernels. Even though these methods have obtained relatively better performances, they are usually time-consuming and not robust for different conditions. To address this problem, we propose a multi-task learning mechanism for remote sensing image motion deblurring in this paper, which contains an image restoration subtask and an image texture complexity recognition one. First, we consider the image motion deblurring problem as a domain transformation problem, from the blurred domain to a clear one. Specifically, the blurred domain represents the data space consisted of blurring images, and the defination of clear domain is similar. Second, we design a novel weighted attention map loss to enhance the reconstruction capability of the restoration subbranch for difficult local regions. Third, based on the restoration subbranch, a recognition subbranch is incorporated into the framework to guide the deblurring process, which provides the auxiliary texture complexity information to help the optimization of restoration subbranch. Additionally, in order to optimize the proposed network, we construct three large-scale datasets, and each sample in the dataset contains a clear image, a blurred image and its texture label obtained by corresponding texture complexity. Finally, the experimental results on three constructed datasets demonstrate the robustness and the effectiveness of the proposed method.