MS-YOLOv7:YOLOv7 Based on Multi-Scale for Object Detection on UAV Aerial Photography

Abstract

A multi-scale UAV aerial image object detection model MS-YOLOv7 based on YOLOv7 was proposed to address the issues of a large number of objects and a high proportion of small objects that commonly exist in the Unmanned Aerial Vehicle (UAV) aerial image. The new network is developed with a multiple detection head and a CBAM convolutional attention module to extract features at different scales. To solve the problem of high-density object detection, a YOLOv7 network architecture combined with the Swin Transformer units is proposed, and a new pyramidal pooling module, SPPFS is incorporated into the network. Finally, we incorporate the SoftNMS and the Mish activation function to improve the network’s ability to identify overlapping and occlusion objects. Various experiments on the open-source dataset VisDrone2019 reveal that our new model brings a significant performance boost compared to other state-of-the-art (SOTA) models. Compared with the YOLOv7 object detection algorithm of the baseline network, the mAP0.5 of MS-YOLOv7 increased by 6.0%, the mAP0.95 increased by 4.9%. Ablation experiments show that the designed modules can improve detection accuracy and visually display the detection effect in different scenarios. This experiment demonstrates the applicability of the MS-YOLOv7 for UAV aerial photograph object detection.

A military reconnaissance network for small‐scale open‐scene camouflaged people detection

“…One‐stage detectors directly regard object location and classification as a regression problem, without the proposed generation steps of the region. The most representative methods are SSD (Ni et al, 2023; Wang, Wang, et al, 2022) and YOLO (Qin et al, 2022; Vajgl et al, 2022; Zhao & Zhu, 2023) families, which perform multi‐scale prediction of targets to narrow the differences between scale distributions. Inspired by a single‐shot multi‐box detector (SSD), SSD‐Like (Ni et al, 2023) proposed an improved dual‐threshold non‐maximum suppression (DT‐NMS) algorithm to alleviate the brightness and occlusion problems caused by complex indoor environments.…”

“…L. Qin et al proposed a real-time salient object detection network ID-YOLO (Qin et al, 2022) to identify the key targets in the driver's gaze area. Because of the superiority of YOLOv7 (Wang, Bochkovskiy, et al, 2022) in accuracy and speed, MS-YOLOv7 (Zhao & Zhu, 2023) is presented to solve the trade-off between accuracy and reasoning speed and has good results in scenes with dense distribution of targets. However, despite the strong development of the above target detection algorithm, it is still an unsolved challenge to detect small-scale targets, especially in the case of camouflage skills.…”

“…To our knowledge, however, there is no specific and open dataset for long-range camouflage object detection in the military domain so far. For the most relevant work (Yi et al, 2021;Zheng et al, 2019), the camouflaged people dataset (Zheng et al, 2019) is an image taken at a normal distance with a large percentage of pixels occupied by the object, whereas the MOD dataset (Yi et al, 2021) is only in the military domain without focusing on camouflage scenarios, where the object prospects are very clearly distinguishable. As such, none of them can be applied to the study of this paper and it is necessary to build a specific dataset for the detection of long-distance disguised people in multiple scenes.…”

“…Some examples from the dataset are shown in Figure 1. In addition, we also re-labelled the object boxes from (Zheng et al, 2019), as they were originally utilized as pixel-level labels for semantic segmentation instead of object detection. Our dataset will be available on the web after acceptance.…”

“…In this open scene, the appearance of camouflage technology makes it more difficult to detect targets. Camouflage targets mainly imitate the same colour, pattern, texture, intensity, and other forms of the background, such as dazzle camouflage (Scott-Samuel et al, 2011), mimesis (Zheng et al, 2019), disruptive colonization (Stevens et al, 2006), and distinctive markings (Dimitrova et al, 2009), thus disguising or hiding the foreground targets in the environment. Under the coexistence of camouflage technology and open scene, because the long-distance target is usually small, the foreground features of the target are more easily covered by complex background noise, which further increases the difficulty of object detection.…”