Aiming at the problem that the existing UAV group target detection algorithm does not consider the connection between group individuals and is prone to miss detection of UAV, this paper proposes a group target detection algorithm for UAV group based on group structure. Based on the group structure information, the Group Structure Relation module is constructed to improve the detection and positioning ability of UAV group targets. Experiments were carried out on the constructed dataset. The experimental results show that the mAP of the proposed UAV group target detection algorithm reaches 91.3%, which is about 2% higher than that of the original YOLOv5 algorithm, effectively improving the detection accuracy of UAV group members. At the same time, the detection speed reaches 65 frames per second, realizing the real-time detection of UAV group targets.
Remote sensing image ship target detection and course discrimination is one of the important supports for building a maritime power. Since ship target in remote sensing images are generally in strips, the IOU score is very sensitive to the angle of bounding box. Moreover, the angle of the ship is a periodic function, this discontinuity will cause performance degeneration. Meanwhile, methods generally use oriented bounding boxes as anchors to handle rotated ship target and thus introduce excessive hyper-parameters such as box size, aspect ratios. Aiming at the problem of complex calculation of anchor frame traversal mechanism and discontinuity of angle regression caused by increasing angle attribute in ship target detection of remote sensing image, a ship target heading detection method based on ship head point is proposed. The discontinuous angle regression problem is transformed into a continuous key point estimation problem, and the ship target detection and heading recognition are unified. Second, CA attention mechanism is added to the feature extraction network to enhance the attention to the ship target and predict the center point of the ship target. The offset and target width at the center point are regressed. Then, return the heading point and offset to obtain the accurate heading point position. Next, the rotation angle of the ship is determined according to the coordinates of the center point and the ship head point. Combined with the predicted width and height of the ship, the rotation frame detection of the ship target is completed. Finally, the center point and the bow point are connected to determine the course of the ship target. The effectiveness of the proposed method is verified on the RFUE and open source HRSC2016 datasets, respectively, and it also has good robustness in complex environments.
The misuse of UAVs has spurred the development of Anti-UAV technology. Infrared detector-based UAV tracking technology has become a research hotspot in the field of the Anti-UAV technology, but still faces the problem of tracking failure caused by background interference. To improve the accuracy and stability of infrared UAV tracking in the complex environments, a spatial-temporal joint constraints based infrared UAV tracking algorithm is proposed. First, a feature pyramid-based Siamese backbone is constructed to enhance the capability of feature extraction for infrared UAVs through cross-scale feature fusion. Next, a region proposal network based on spatio-temporal joint constraints is proposed. Under the constraints of template appearance features and target motion information, the location probability distribution of the infrared UAV is predicted in the entire image, and the prior anchor box is guided to focus on the candidate regions, realizing a soft adaptive search region selection mechanism. By focusing the search area, the anti-background interference capability of the local search strategy and the recapture capability of global search strategy are fused, which effectively mitigates the negative sample interference brought by global search and further enhances the discriminability of target features. Finally, the proposed algorithm is evaluated on the Anti-UAV dataset, achieving precision, success rate, and average precision of 89.5%, 64.9%, and 65.6%, respectively, with a tracking speed of 18.5 FPS. Compared with other advanced tracking algorithms, the proposed algorithm obtains better tracking performance and superior tracking performance in complex scenarios such as fast motion, thermal crossover and distractors interference.
Currently, infrared time-sensitive target detection technology is widely used in military and civil applications such as air defense and early warning, maritime surveillance, and precision guidance, but some high-value target images are difficult and expensive to acquire. To address the problems such as the lack of infrared time-sensitive target image data and the lack of multi-scene multi-target data for training, this paper proposes an infrared time-sensitive target data enhancement algorithm based on a generative model, which is a two-stage model. Firstly, in the first stage, the visible images containing time-sensitive targets are converted to infrared images by a modal conversion model based on CUT networks. Then in the second stage a large number of random targets are generated from the converted IR images using an adversarial random sample generation model to achieve the data enhancement effect. The coordinate attention mechanism is also introduced into the generator module in the second stage, which effectively enhances the feature extraction capability of the network. Finally, modal conversion experiments and sample random generation experiments are conducted, and the results show the feasibility of the data enhancement method of generative model proposed in this paper in IR time-sensitive target data enhancement, which provides a strong data support for improving IR time-sensitive target detection algorithm.
To address the problems of complex background of land buildings and islands in near-shore SAR image ship detection, dense ship docking, and thus inaccurate localization and target miss detection, we propose a YOLOv7 near-shore SAR ship rotation target detection model based on the attention mechanism and KLD improvement. Firstly, considering the lack of attention mechanism and remote dependency of YOLOv7, CA attention mechanism is added to the backbone network to improve the model context encoding capability and enhance the model accuracy. Secondly, the 3D nonreference attention mechanism SimAm is introduced to further improve the attention to ship features. Finally, the angular information is considered for the problem that the ship targets of SAR images are closely aligned in any direction. KLD is used as the localization loss function. The experimental results on the SSDD dataset show that the improved algorithm in this paper improves AP by 14.34% in near-shore scenes and the same in offshore scenes, with 2.22% improvement in all scenes relative to the original YOLOv7 model. The experimental results show that the algorithm applies to detecting ship targets in any direction in the near-shore scenes.
Heterogeneous image matching is a hot and difficult research topic in the field of image processing. The existing visible and infrared image matching has problems such as large modal differences, difficult matching, and poor robustness. Therefore, an intelligent matching method for visible and infrared images based on BCE-CycleGAN is proposed. First, a BCECycleGAN model is proposed based on the image style translation with generative adversarial networks, it can convert visible images to infrared images. By designing a new generative network loss function, the transformation effect of the model on heterogeneous images is improved. Then, the generated infrared images are matched with the original infrared images using LoFTR and DFM algorithms. LoFTR and DFM are currently advanced deep learning-based intelligent matching algorithms. Finally, the conversion relationship is mapped to the corresponding visible and infrared image pair to obtain the final matching result. Images style translation experiments and matching experiments on the test datasets show that the BCE-CycleGAN network proposed in this paper can effectively reduce the complexity of the algorithm and improve the quality of image generation. Furthermore, combining BCE-CycleGAN with deep learning-based matching methods can effectively improve the effectiveness and robustness of the matching algorithm.
Aiming at the problem that the traditional scene matching navigation algorithm needs to manually design features, a scene matching navigation algorithm based on improved Siamese network is proposed. First, the space transformation network module is fused with the original Siamese network to improve the fitting ability between the scene features, and then the improved network is applied to the location and orientation algorithm of aircraft. The experimental results show that the Siamese network image matching navigation algorithm based on the fusion space transformation module enhances the ability to deal with the rotation and translation transformation between the real-time image and the reference image. Compared with the original Siamese network, the algorithm in this paper optimizes the similarity of two scenes from different angles by an average of 9.04%, thus expanding the adaptability of the algorithm. Compared with the traditional template matching algorithm, this algorithm has higher matching accuracy and stronger robustness when the angle of the real-time image changes and has certain practical application value in navigation algorithms.
Laser active imaging is widely used in many fields. The intensity image quality of laser active imaging is affected by various degradations, such as speckle effect and noise. Most existing objective image quality assessment (IQA) methods that consider only a single distorted image are not suitable for an intensity image. A multiscale full-reference intensity IQA method is presented. The proposed method is based on an improved gradient magnitude similarity deviation (GMSD) in the nonsubsampled contourlet transform (NSCT) domain. The reference and distorted images are decomposed by NSCT to emulate the multichannel structure of the human visual system. Then, the GMSD of each sub-band is computed to capture the intensity image quality. At last, the contrast sensitivity function implementation is employed in the sub-bands of the NSCT domain. All sub-bands’ GMSD is evaluated and pooled together to yield the objective quality index of a distorted intensity image. Experimental results show that the proposed method can effectively and accurately evaluate the quality of intensity images, and it is highly consistent with subjective perception.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.