Search

ObjectiveThis paper proposes a lightweight remote sensing ship target detection algorithm LR-YOLO based on improved YOLOv5s to meet the lightweight and fast inference requirements of ship target detection tasks involving remote sensing images. MethodsFirst, the backbone network adopts the ShuffleNet v2 block stacking method, effectively reducing the number of network model parameters and improving the computational speed; second, a region selection module filter is designed to select regions of interest and extract effective features more fully; finally, a circular smooth label is introduced to calculate angle loss and perform rotation detection on remote sensing ship targets, while deformable convolution is used to adapt to geometric deformation and improve detection performance. ResultsThe experimental results on the HRSC2016 ship dataset show that the detection accuracy of the algorithm reaches 92.90%, an improvement of 1.3%, with the number of network model parameters only 39.33% that of the baseline model. ConclusionThe proposed algorithm achieves a balance between lightweight and detection accuracy, providing references for remote sensing ship target detection.

Deep learning techniques have made significant advancements in computer vision. The YOLO algorithm, a representative single-stage detection approach, has demonstrated remarkable results in detecting ship targets in SAR images. We introduce an enhanced ship target detection model for SAR images, utilizing the improved YOLOv7 object detection network. We incorporate the coordinate attention mechanism into the network to enable automatic detection and diagnosis of ship targets within SAR images. To enhance the robustness and positioning accuracy of the detection network, we replace the CIoU regression loss in YOLOv7 with the SIoU loss, reducing the complexity of the loss function. Additionally, we integrate the rotating target detection technology into the network to mitigate the impact of target overlap on detection results. Comprehensive experiments conducted on the Capella Space synthetic aperture radar datasets validate that the proposed methodology achieves superior performance in multiple evaluation metrics, including precision, recall, and mean average precision. [ABSTRACT FROM AUTHOR]

This paper proposes a multi-scale information fusion based remote sensing small target detection method that aims to address the issue of low target detection accuracy in optical remote sensing images due to complex backgrounds, diversified scales, small targets, and different directions. Firstly, the architecture of the RepConv module significantly increases the detection accuracy of small targets without adding more inference time. Secondly, by introducing the ECA attention mechanism, a C3ECA module is constructed to effectively reduce the interference of complex background areas and achieve accurate positioning of the target area. The PANet structure in YOLOv5 is replaced by the BiFPN structure to balance the feature information of different scales and improve the detection performance of multi-scale objects. In addition, in order to solve the uncertainty of target direction and reduce the boundary discontinuity caused by angle regression, a circular smooth label method is used to provide an effective solution for target detection. The preprocessing method of image slices is employed to successfully achieve the target detection of high-resolution images. This approach greatly minimizes the issue of missed detection and erroneous detection of small objects in large images. The experimental findings indicate that the proposed approach markedly enhances the accuracy of remote sensing image detection and offers notable benefits in the recognition of small-scale targets. [ABSTRACT FROM AUTHOR]

Aiming at the problems that the current robot grasping detection method is too discrete in predicting the grasping angle and the grasping process may produce large off-angle, which reduces the grasping detection accuracy and even leads to grasping failure, an improved robot real- time grasping detection method based on the YOLOv5 neural network model is proposed. Firstly, the grasping frame coordinates and grasping angles are extracted based on the single-stage object detection model YOLOv5. Afterwards, the grasping angles are divided more carefully, while circular smoothing label is introduced to accommodate the periodicity of the angles, links between adjacent angles are established, the YOLOv5 detection head is decoupled, and the loss function is optimized to improve the detection accuracy. Finally, an experimental validation is performed on the Cornell dataset. The experimental results show that the proposed algorithm can better predict the grasping angle and improve the grasping detection accuracy compared with the classical grasping detection methods. The model achieves 97.5% accuracy and 71 FPS detection speed on the Cornell dataset. [ABSTRACT FROM AUTHOR]

Secure grasping of objects in complex scenes is the foundation of many tasks. It is important for robots to autonomously determine the optimal grasp based on visual information, which requires reasoning about the stacking relationship of objects and detecting the grasp position. This paper proposes a multi-task secure grasping detection model, which consists of the grasping relationship network (GrRN) and the oriented rectangles detection network CSL-YOLO, which uses circular smooth label (CSL). GrRN uses DETR to solve set prediction problems in object detection, enabling end-to-end detection of grasping relationships. CSL-YOLO uses classification to predict the angle of oriented rectangles, and solves the angle distance problem caused by classification. Experiments on the Visual Manipulate Relationship Dataset (VMRD) and the grasping detection dataset Cornell demonstrate that our method outperforms existing methods and exhibits good applicability on robot platforms. [ABSTRACT FROM AUTHOR]

Wheat impurity content is one of the crucial indicators for evaluating wheat quality grades and prices, serving as an essential inspection parameter for grain storage. Due to the substantial intra-class variations in the aspect ratios of wheat impurities, traditional object detection algorithms fail to consider the diverse shapes and sizes of wheat impurities in real-world scenarios, resulting in low detection accuracy and missed detections. To address those issues, a YOLOv5-based wheat impurity automatic detection method, named Ro-YOLOv5, is proposed for computed tomography (CT) images of wheat samples. This method introduces rotated detection boxes and Circular Smooth Label (CSL) to transform angle regression predictions into angle classification predictions, effectively resolving issues related to angle periodicity and boundary exchange in regression; Additionally, the GIOU loss function in YOLOv5 is replaced with the CIOU loss function, which is more suitable for wheat impurity targets, ensuring the lower loss values and stable boundary regression. When creating the impurity-containing wheat dataset, a hybrid filtering method is introduced for leveraging the damping characteristics of the S-L filter for suppressing high-frequency components in projection images, thereby mitigating oscillation phenomena; Moreover, the Blackman filter is employed for image sharpening to preserve more image detail components. Experiments results have shown that the proposed algorithm achieves a mean average precision of 88.83% on the self-made dataset, representing a 7.56% improvement over the original YOLOv5 baseline network. The average detection time is 11.6ms, meeting the performance requirements for detecting wheat impurities. On the basis of the traditional YOLOv5 object detection network, a Ro-YOLOv5 network is proposed for the impurity detection in wheats CT images, taking into account the geometric characteristics of wheat impurities. The following innovative works are involved. • Considering the geometric features of wheat impurities, the CSL representation method is introduced into the traditional YOLOv5 model, aiming to transform angle regression predictions into angle classification predictions, effectively resolving issues related to angle periodicity and boundary exchange in regression. • The rotation smooth labels with angle information replaces the horizontal labels in the raditional YOLOv5 network, the geometric and angular periodicity issues related to wheat impurities are addressed, and ineffective annotation areas of bounding boxes are reduced, thereby reducing the computational complexity of the network. • The boundary loss function is replaced with the CIoU loss function, which is better suited for the geometric characteristics of wheat impurities. The experiments are developed and have demonstrated that the proposed Ro-YOLOv5 is capable of detecting impurities in wheat CT images, with the higher recognition accuracy and the better real-time capability. [ABSTRACT FROM AUTHOR]

• Creating a maize leaf image dataset with orientation details using a customized UAV-based phenotyping platform. • Developing an automated pipeline for precise detection and quantification of leaf azimuth angle in large-scale maize plants. • The proposed method effectively tackles the challenges of detecting dense and overlapping maize leaves in the field. Maize is the most productive grain crop in the world in terms of total production and yield. Leaf azimuth angle, the directional orientation of the maize leaf, is an important characteristic that affects the efficiency of sunlight reception and photosynthesis in maize canopies. To study maize plant architecture, most of the previous studies have been focusing on leaf vertical angle characterization due to the inherent difficulty of measuring leaf azimuth angle accurately. Achieving accurate and efficient measurement of leaf azimuth angle in maize is important for yield improvement. Traditionally, leaf azimuth angle is measured manually using a protractor or manual annotations of images, which is time-consuming, costly, and prone to human error. Recent advances in Unmanned Aerial Vehicle (UAV) and artificial intelligence have shown great potential in leaf azimuth angle detection. However, most of the deep learning algorithms for object detection are developed for Horizontal Bounding Boxes (HBBs). To measure the azimuth angle of maize leaves, a robust Oriented Bounding Boxes (OBBs) detection algorithm is needed. In this study, a dataset containing top-view images of maize plants was collected using an UAV. An automated tool, Swin-Roleaf, was developed for maize leaf detection. It is an improvement of the YOLOv5 model, uniquely integrating the Swin Transformer (ST) and Circular Smooth Label (CSL). Swin-Roleaf was able to accurately detect OBBs of individual leaves and thereby enable the estimation of leaf azimuth angle. The experimental results showed that Swin-Roleaf was able to accurately detect maize leaves and measure the azimuth angle with a model OBBAP 50 of 90.59% on OBBs. The system-derived leaf azimuth angle was highly correlated with the ground truth with a correlation coefficient larger than 0.96. Our study demonstrates for the first time the feasibility of using UAVs coupled with deep learning algorithms to detect leaf azimuth angles under field conditions. The system can significantly speed up the measurement of maize leaf azimuth angles and provide a new tool to accelerate breeding programs. [ABSTRACT FROM AUTHOR]

Secure grasping of objects in complex scenes is the foundation of many tasks. It is important for robots to autonomously determine the optimal grasp based on visual information, which requires reasoning about the stacking relationship of objects and detecting the grasp position. This paper proposes a multi-task secure grasping detection model, which consists of the grasping relationship network (GrRN) and the oriented rectangles detection network CSL-YOLO, which uses circular smooth label (CSL). GrRN uses DETR to solve set prediction problems in object detection, enabling end-to-end detection of grasping relationships. CSL-YOLO uses classification to predict the angle of oriented rectangles, and solves the angle distance problem caused by classification. Experiments on the Visual Manipulate Relationship Dataset (VMRD) and the grasping detection dataset Cornell demonstrate that our method outperforms existing methods and exhibits good applicability on robot platforms.

A novel lightweight Rotational Object Detection algorithm is proposed to overcome the shortcomings of conventional computer-vision-aided object detection methods used in Remote Sensing and Surveillance that overlook the variability in size and orientation of objects in satellite and aerial images. This advanced algorithm integrates a branch dedicated to angle prediction and employs the circular smooth label (CSL) method for angle classification. This approach is suitable for scenarios that require detection in rotational boxes. Our work is further distinguished by the introduction of a novel Channel and Spatial Attention (CSA) module, which is seamlessly integrated into the YOLOv5-CSL framework via the C3CS module. This module accentuates pertinent features through both the channel and spatial attention mechanisms. In addition, bicubic interpolation and the GELU activation function were incorporated into the YOLOv5-CSLA model. Our model achieved 57.86 mAP on the challenging DOTA v2 dataset surpassing the second-best method by 0.20 points and simultaneously consuming 11 million fewer parameters and 103 fewer GFLOPs (our model consumes 25 M Params and 54 GFLOPs), justifying its suitability for deployment on a large majority of platforms, as the compute required is a challenge in real-time deployment scenarios. • Integration of channel and spatial attention in YOLOv5-CSL. • Integration of circular smooth labels and BCE with logit loss using YOLOv5. • The model achieved an mAP of 57.86 on the DOTA-v2 satellite images dataset. • The model consumes just 54 GFLOPS. [ABSTRACT FROM AUTHOR]

Arbitrary-oriented object detection has been a building block for rotation sensitive tasks. We first show that the boundary problem suffered in existing dominant regression-based rotation detectors, is caused by angular periodicity or corner ordering, according to the parameterization protocol. We also show that the root cause is that the ideal predictions can be out of the defined range. Accordingly, we transform the angular prediction task from a regression problem to a classification one. For the resulting circularly distributed angle classification problem, we first devise a Circular Smooth Label technique to handle the periodicity of angle and increase the error tolerance to adjacent angles. To reduce the excessive model parameters by Circular Smooth Label, we further design a Densely Coded Labels, which greatly reduces the length of the encoding. Finally, we further develop an object heading detection module, which can be useful when the exact heading orientation information is needed e.g. for ship and plane heading detection. We release our OHD-SJTU dataset and OHDet detector for heading detection. Extensive experimental results on three large-scale public datasets for aerial images i.e. DOTA, HRSC2016, OHD-SJTU, and face dataset FDDB, as well as scene text dataset ICDAR2015 and MLT, show the effectiveness of our approach. [ABSTRACT FROM AUTHOR]