Object Detection Approach Research Articles

Behavioral response of fish under ammonia nitrogen stress based on machine vision

The long-term accumulation of ammonia nitrogen in aquaculture seriously affects the life of fish and even causes large-scale death. Moreover, when the concentration of ammonia nitrogen starts to accumulate, it is a judgment standard to provide early warning through the changes in fish behavior to prevent excessive ammonia nitrogen in water. Therefore, this paper proposes a novel approach to monitoring water quality for aquaculture based on deep learning and three-dimensional movement trajectory. The improved YOLOv8 model was used as the object detection approach to obtain three-dimensional position information of fish by combining Kalman filter, Kuhn Munkres (KM) algorithm, and Kernelized Correlation Filters (KCF) algorithm. The proposed approach was evaluated in the recovery experiment of acute ammonia nitrogen stress of sturgeon, bass, and crucian. The experimental results show that the precision, recall, mAP@0.5, and mAP@0.5:0.95 of the improved YOLOv8 model are 0.964, 0.914, 0.979, and 0.602, respectively. In addition, the proposed three-dimensional positioning approach can qualitatively and quantitatively analyze the fish behavior in different stages and further explores the fish behavior changes through behavior trajectories, volumes of exercise, spatial distribution, and movement velocity. This research provides a new method and idea for studying the abnormal behavior of aquatic animals under ammonia nitrogen stress and has theoretical and practical significance.

ELWNet: An Extremely Lightweight Approach for Real-Time Salient Object Detection

Existing lightweight salient object detection (SOD) methods aim to solve the problem of high computational costs that is prevalent with heavyweight methods. However, compared with heavyweight methods, the detection accuracy of lightweight methods is greatly reduced while real-time performance is not significantly improved. Therefore, we aim to establish a trade off between computational cost and detection performance by improving the network efficiency. We propose a fast and extremely lightweight end-to-end wavelet neural network (ELWNet) for real-time salient object detection. ELWNet can achieve salient object detection and segmentation at approximately 70FPS (GPU), 19FPS (CPU) with 76K parameters and 0.38G FLOPs. We introduce wavelet transform theory into a neural network, proposing a wavelet transform module (WTM), a wavelet transform fusion module (WTFM), a novel feature residual mechanism, and construct an efficient architecture. The wavelet transform theory is integrated into the neural network to realize the interaction between the features in the frequency and the time domain. Meanwhile, ELWNet does not rely on a pre-trained model, which significantly reduces redundant features. We validate the performance of ELWNet using five well-known datasets, and demonstrate state-of-the-art performance compared with 24 other SOD models in terms of being lightweight, detection accuracy and real-time capabilities. Our method maintains high detection performance while reducing the number of model parameters by approximately 99% compared with heavyweight methods.

CerviCell-detector: An object detection approach for identifying the cancerous cells in pap smear images of cervical cancer

Cervical cancer is the second most commonly seen cancer in women. It affects the cervix portion of the vagina. The most preferred diagnostic test required for screening for cervical cancer is the pap smear test. Pap smear is a time-consuming test as it requires detailed analysis by expert cytologists. Cytologists can screen around 100 to 1000 slides depending upon the availability of advanced equipment. It requires substantial time and effort to carefully examine each slide, identify and classify cells, and make accurate diagnoses. Prolonged periods of visual inspection can increase the likelihood of human errors, such as overlooking abnormalities or misclassifying cells. The sheer volume of slides to be screened can exacerbate fatigue and impact diagnostic accuracy. Due to this reason Artificial intelligence (AI) based computer-aided diagnosis system for the classification and detection of pap smear images is needed. There are some AI-based solutions proposed in the literature, still, an effective and accurate system is under research. In this paper, we implement a state-of-the-art object detection model with a newly available CRIC dataset which follows the Bethesda system for nomenclature. Object detection models implemented are YOLOv5 which uses the CSPNet backbone, Faster R–CNN which has Region Proposal Network (RPN) and Detectron2 framework created by Facebook AI Research (FAIR) Group. ResNext model is implemented among the available models from Detectron2. The CRIC dataset is preprocessed and augmented using Roboflow tool. The performance measures of Average Precision and mean Average precision over the Intersection over Union (IoU) are used to evaluate the effectiveness of the models. The models performed better for two classes namely Normal and Abnormal compared to six classes from the Bethesda system. The highest mean Average Precision (mAP) is observed on the augmented dataset for YOLOv5 models for binary classification with 83 % mAP with IoU in the range of 0.50–0.95.

Enhancement of Real-Time Object Detection and Tracking in Collaborative Environment using AI and Mixed Reality

The area of mixed reality has had rapid growth in recent years, with a notable rise in funding. This may be attributed to the rising recognition of the potential advantages associated with the integration of virtual information into the physical environment. The majority of contemporary mixed reality apps that rely on markers use algorithms for local feature identification and tracking. This study aims to enhance the accuracy of object recognition in complicated environment and enable real-time classification operations via the introduction of a unique detection approach known as the lightweight and efficient YOLOv4 model. In the present setting, Computational vision emerges as a very valuable and engaging manifestation of artificial intelligence (AI) that finds widespread application in many aspects of daily existence. The field of computer vision is dedicated to the development of advanced artificial intelligence and computer systems that aim to replace complex elements of the human environment. In recent times, deep neural networks have emerged as a crucial component in several sectors owing to their well-established capacity to process visual input. This study presents a methodology for classifying and identifying objects using the YOLOv4 object detection algorithm. Convolutional neural networks (CNNs) have shown exceptional efficacy in the tasks of object tracking and feature extraction from pictures. Therefore, the enhanced network architecture optimizes both the precision of identification and the speed at which it operates. This research will contribute to developing mixed-reality simulations system for object detection and tracking in collaborative environment that are accessible to everyone, including users in the architectural filed. The model was evaluated in comparison to other object detection approaches. Based on the empirical results, it was observed that the YOLOv4 model exhibited a mean average precision (mAP) of 0.988, surpassing the performance of both YOLOv3 and other object identification models.

Sniffer Faster R-CNN ++: An Efficient Camera-LiDAR Object Detector with Proposal Refinement on Fused Candidates

In this paper we present Sniffer Faster R-CNN++, an efficient Camera-LiDAR late fusion network for low complexity and accurate object detection in autonomous driving scenarios. The proposed detection network architecture operates on output candidates of any 3D detector and proposals from regional proposal network of any 2D detector to generate final prediction results. In comparison to the single modality object detection approaches, fusion based methods in many instances suffer from dissimilar data integration difficulties. On one hand, fusion based network models are complicated in nature and on the other hand they require large computational overhead and resources, processing pipelines for training and inference specially, the early fusion and deep fusion approaches. As such, we devise a late fusion network that in-cooperates pre-trained, single-modality detectors without change, performing association only at the detection level. In addition to this, lidar based method fail to detect distant object due to its sparse nature so we devise proposal refinement algorithm to jointly optimize detection candidates and assist detection for distant objects. Extensive experiments on both the 3D and 2D detection benchmark of challenging KITTI dataset illustrate that our proposed network architecture significantly improves the detection accuracy, accelerating the detection speed.

Performance Comparison of Sea Cucumber Detection by the Yolov5 and DETR Approach

Sea cucumber detection represents an important step in underwater environmental perception, which is an indispensable part of the intelligent subsea fishing system. However, water turbidity decreases the clarity of underwater images, presenting a challenge to vision-based underwater target detection. Therefore, accurate, real-time, and lightweight detection models are required. First of all, the development of subsea target detection is summarized in this present work. Object detection methods based on deep learning including YOLOv5 and DETR, which are, respectively, examples of one-stage and anchor-free object detection approaches, have been increasingly applied in underwater detection scenarios. Based on the state-of-the-art underwater sea cucumber detection methods and aiming to provide a reference for practical subsea identification, adjacent and overlapping sea cucumber detection based on YOLOv5 and DETR are investigated and compared in detail. For each approach, the detection experiment is carried out on the derived dataset, which consists of a wide variety of sea cucumber sample images. Experiments demonstrate that YOLOv5 surpasses DETR in low computing consumption and high precision, particularly in the detection of small and dense features. Nevertheless, DETR exhibits rapid development and holds promising prospects in underwater object detection applications, owing to its relatively simple architecture and ingenious attention mechanism.

Infield corn kernel detection using image processing, machine learning, and deep learning methodologies under natural lighting

Machine vision has been increasingly used to address agricultural issues. One such case is corn field harvest losses and image-based object detection approaches, namely image processing, machine learning, and deep learning were investigated to detect and count infield corn kernels, immediately after harvest for combine harvester performance evaluation. A hand-held low-cost RGB camera was used to collect images with kernels of different backgrounds, based on which a 420 images dataset (200, 40, and 180 for training, validation, and testing, respectively) was generated. Three different models for kernel detection were constructed based on image processing, machine learning, and deep learning. For the imaging processing method, the images were preprocessed (color thresholding, graying, and erosion), followed by Hough circle detection to identify kernels. For the machine learning (cascade detector) and deep learning (Mask R-CNN, EfficientDet, YOLOv5, and YOLOX), models were trained, validated, and tested. Experimental results showed the overall performance of the deep learning network YOLOv5 was superior to the other approaches, with a small model size (89.3 MB) and a high model average precision (78.3 %) for object detection. The detection accuracy, undetection rate and F1 value were 90.7 %, 9.3 %, and 91.1 %, respectively, and the average detection rate was 55 fps. This study demonstrates that the YOLOv5 model has the potential to be used as a real-time, reliable, and robust method for infield corn kernel detection.

Multi-Branch Dilation Convolution CenterNet for Object Detection of Underwater Vehicles

Object detection occupies a very important position in the fishing operation and autonomous navigation of underwater vehicles. At present, most deep-learning object detection approaches, such as R-CNN, SPPNet, R-FCN, etc., have two stages and are based on anchors. However, the previous methods generally have the problems of weak generalization ability and not high enough computational efficiency due to the generation of anchors. As a well-known one-stage anchor-free method, CenterNet can accelerate the inference speed by omitting the step of generating anchors, whereas it is difficult to extract sufficient global information because of the residual structure at the bottom layer, which leads to low detection precision for the overlapping targets. Dilation convolution makes the kernel obtain a larger reception field and access more information. Multi-branch structure can not only preserve the whole area information, but also efficiently separate foreground and background. By combining the dilation convolution and multi-branch structure, multi-branch dilation convolution is proposed and applied to the Hourglass backbone network in CenterNet, then an improved CenterNet named multi-branch dilation convolution CenterNet (MDC-CenterNet) is built, which has a stronger ability of object detection. The proposed method is successfully utilized for detection of underwater organisms including holothurian, scallop, echinus and starfish, and the comparison result shows that it outperforms the original CenterNet and the classical object detection network. Moreover, with the MS-COCO and PASCAL VOC datasets, a number of comparative experiments are performed for showing the advancement of our method compared to other best methods.

ExistenceMap-PointPillars: A Multifusion Network for Robust 3D Object Detection with Object Existence Probability Map.

Recognition of surrounding objects is crucial for ensuring the safety of automated driving systems. In the realm of 3D object recognition through deep learning, several methods incorporate the fusion of Light Detection and Ranging (LiDAR) and camera data. The effectiveness of the LiDAR-camera fusion approach is widely acknowledged due to its ability to provide a richer source of information for object detection compared to methods that rely solely on individual sensors. Within the framework of the LiDAR-camera multistage fusion method, challenges arise in maintaining stable object recognition, especially under adverse conditions where object detection in camera images becomes challenging, such as during night-time or in rainy weather. In this research paper, we introduce "ExistenceMap-PointPillars", a novel and effective approach for 3D object detection that leverages information from multiple sensors. This approach involves a straightforward modification of the LiDAR-based 3D object detection network. The core concept of ExistenceMap-PointPillars revolves around the integration of pseudo 2D maps, which depict the estimated object existence regions derived from the fused sensor data in a probabilistic manner. These maps are then incorporated into a pseudo image generated from a 3D point cloud. Our experimental results, based on our proprietary dataset, demonstrate the substantial improvements achieved by ExistenceMap-PointPillars. Specifically, it enhances the mean Average Precision (mAP) by a noteworthy +4.19% compared to the conventional PointPillars method. Additionally, we conducted an evaluation of the network's response using Grad-CAM in conjunction with ExistenceMap-PointPillars, which exhibited a heightened focus on the existence regions of objects within the pseudo 2D map. This focus resulted in a reduction in the number of false positives. In summary, our research presents ExistenceMap-PointPillars as a valuable advancement in the field of 3D object detection, offering improved performance and robustness, especially in challenging environmental conditions.

A novel deep learning based underwater image de-noising and detecting suspicious object

Exploration of underwater resource play a vital role for nation development. Underwater surveillance systems play a crucial role in security applications, requiring accurate detection of suspicious objects in underwater images. However, the presence of noise, poor visibility, and uneven lighting conditions in underwater environments pose significant challenges for reliable object detection. This work proposes an integrated approach for underwater image de-noising, pre-processing, enhancement, and subsequent suspicious object detection by combining the DnCNN (Deep Convolutional Neural Network), CLAHE (Contrast Limited Adaptive Histogram Equalization), and additional image enhancement techniques. In addition to de-noising and pre-processing, it incorporate various image enhancement techniques to further improve object detection performance. These techniques include color correction, contrast adjustment, and edge enhancement, aiming to enhance the visual characteristics and saliency of suspicious objects in underwater images. To evaluate the effectiveness of proposed approach, this work conducted extensive experiments on an underwater image dataset containing diverse scenes and suspicious objects. The work compares proposed method with existing de-noising, preprocessing, and object detection techniques, analyzing the results using quantitative performance metrics, including precision, recall, and F1 score. The experimental results demonstrate that proposed integrated approach outperforms individual methods and achieves superior detection performance by enhancing the quality of underwater images and improving the visibility of suspicious objects.

Saliency-Induced Moving Object Detection for Robust RGB-D Vision Navigation Under Complex Dynamic Environments

Localization in unknown environments is an essential requirement for vision navigation of robotic vehicles in intelligent transportation systems. However, moving objects in dynamic scenarios usually bring about great difficulty for robot localization, because motion estimation of robotic vehicles is disturbed by increasing feature outliers caused by moving objects. In order to improve the accuracy and robustness of robot localization, a novel saliency-induced moving object detection (SMOD) approach is proposed to filter out feature outliers for RGB-D-based simultaneous localization and mapping (SLAM) in complex dynamic workspaces. Firstly, three complementary motion saliency potentials, including motion energy (ME), spatiotemporal objectness (STO), and dynamic superpixels (DS), are modeled by fully analyzing spatial, temporal, appearance and depth cues in RGB-D inputs. They can be used to identify the dynamic objects effectively from diversely changing backgrounds. Then, a superpixel-level graph-based motion saliency (MS) measure is proposed to generate the MS map for reliable localization of the moving objects. The edge weights and background nodes on the graph are determined reasonably by fusing ME, STO, and DS, which is not vulnerable to background interferences. Furthermore, the SMOD approach is embedded into the front-end of ORB-SLAM3 as a pre-processing stage, in order to filter out feature outliers associated with the moving objects. Finally, the extensive experiments are performed to verify the accuracy and robustness of the proposed approach on the public dynamic datasets. The experimental results show that the SMOD method can detect the moving objects effectively in a variety of challenging dynamic environments, and separate the dynamic regions reliably from the irrelevant background. The data comparison demonstrates that the SMOD-SLAM navigation system can outperform other state-of-the-art dynamic visual SLAM (vSLAM) systems.

MR Object Identification and Interaction

The increasing number of objects in ubiquitous computing environments creates a need for effective object detection and identification mechanisms that permit users to intuitively initiate interactions with these objects. While multiple approaches to such object detection -- including through visual object detection, fiducial markers, relative localization, or absolute spatial referencing -- are available, each of these suffers from drawbacks that limit their applicability. In this paper, we propose ODIF, an architecture that permits the fusion of object situation information from such heterogeneous sources and that remains vertically and horizontally modular to allow extending and upgrading systems that are constructed accordingly. We furthermore present BLEARVIS, a prototype system that builds on the proposed architecture and integrates computer-vision (CV) based object detection with radio-frequency (RF) angle of arrival (AoA) estimation to identify BLE-tagged objects. In our system, the front camera of a Mixed Reality (MR) head-mounted display (HMD) provides a live image stream to a vision-based object detection module, while an antenna array that is mounted on the HMD collects AoA information from ambient devices. In this way, BLEARVIS is able to differentiate between visually identical objects in the same environment and can provide an MR overlay of information (data and controls) that relates to them. We include experimental evaluations of both, the CV-based object detection and the RF-based AoA estimation, and discuss the applicability of the combined RF and CV pipelines in different ubiquitous computing scenarios. This research can form a starting point to spawn the integration of diverse object detection, identification, and interaction approaches that function across the electromagnetic spectrum, and beyond.

Comparison of Machine Learning Pixel-Based Classifiers for Detecting Archaeological Ceramics

Recent improvements in low-altitude remote sensors and image processing analysis can be utilised to support archaeological research. Over the last decade, the increased use of remote sensing sensors and their products for archaeological science and cultural heritage studies has been reported in the literature. Therefore, different spatial and spectral analysis datasets have been applied to recognise archaeological remains or map environmental changes over time. Recently, more thorough object detection approaches have been adopted by researchers for the automated detection of surface ceramics. In this study, we applied several supervised machine learning classifiers using red-green-blue (RGB) and multispectral high-resolution drone imageries over a simulated archaeological area to evaluate their performance towards semi-automatic surface ceramic detection. The overall results indicated that low-altitude remote sensing sensors and advanced image processing techniques can be innovative in archaeological research. Nevertheless, the study results also pointed out existing research limitations in the detection of surface ceramics, which affect the detection accuracy. The development of a novel, robust methodology aimed to address the “accuracy paradox” of imbalanced data samples for optimising archaeological surface ceramic detection. At the same time, this study attempted to fill a gap in the literature by blending AI methodologies for non-uniformly distributed classes. Indeed, detecting surface ceramics using RGB or multi-spectral drone imageries should be reconsidered as an ‘imbalanced data distribution’ problem. To address this paradox, novel approaches need to be developed.

A deep reinforcement learning assisted task offloading and resource allocation approach towards self-driving object detection

With the development of communication technology and mobile edge computing (MEC), self-driving has received more and more research interests. However, most object detection tasks for self-driving vehicles are still performed at vehicle terminals, which often requires a trade-off between detection accuracy and speed. To achieve efficient object detection without sacrificing accuracy, we propose an end–edge collaboration object detection approach based on Deep Reinforcement Learning (DRL) with a task prioritization mechanism. We use a time utility function to measure the efficiency of object detection task and aim to provide an online approach to maximize the average sum of the time utilities in all slots. Since this is an NP-hard mixed-integer nonlinear programming (MINLP) problem, we propose an online approach for task offloading and resource allocation based on Deep Reinforcement learning and Piecewise Linearization (DRPL). A deep neural network (DNN) is implemented as a flexible solution for learning offloading strategies based on road traffic conditions and wireless network environment, which can significantly reduce computational complexity. In addition, to accelerate DRPL network convergence, DNN outputs are grouped by in-vehicle cameras to form offloading strategies via permutation. Numerical results show that the DRPL scheme is at least 10% more effective and superior in terms of time utility compared to several representative offloading schemes for various vehicle local computing resource scenarios.

Multi-task Learning with Hierarchical Guidance for Locating and Stratifying Submucosal Tumors.

Locating and stratifying the submucosal tumor of the digestive tract from endoscopy ultrasound (EUS) images are of vital significance to the preliminary diagnosis of tumors. However, the above problems are challenging, due to the poor appearance contrast between different layers of the digestive tract wall (DTW) and the narrowness of each layer. Few of existing deep-learning based diagnosis algorithms are devised to tackle this issue. In this article, we build a multi-task framework for simultaneously locating and stratifying the submucosal tumor. And considering the awareness of the DTW is critical to the localization and stratification of the tumor, we integrate the DTW segmentation task into the proposed multi-task framework. Except for sharing a common backbone model, the three tasks are explicitly directed with a hierarchical guidance module, in which the probability map of DTW itself is used to locally enhance the feature representation for tumor localization, and the probability maps of DTW and tumor are jointly employed to locally enhance the feature representation for tumor stratification. Moreover, by means of the dynamic class activation map, probability maps of DTW and tumor are reused to enforce the stratification inference process to pay more attention to DTW and tumor regions, contributing to a reliable and interpretable submucosal tumor stratification model. Additionally, considering the relation with respect to other structures is beneficial for stratifying tumors, we devise a graph reasoning module to replenish non-local relation knowledge for the stratification branch. Experiments on a Stomach-Esophagus and an Intestinal EUS dataset prove that our method achieves very appealing performance on both tumor localization and stratification, significantly outperforming state-of-the-art object detection approaches.

Real Time Object Detection using Deep Learning and SSD Algorithm

Real time object detection is a complex area of computer vision, vast and vibrant. If there is a single object to be detected in an image, it is known as Image Localization and if there are multiple objects in an image, then it is Object Detection. Real time object identification to find and recognize real objects such as cars, bicycles, TV, flowers and people from pictures or videos. Object identification technique allows you to find out image or video details, as it allows detection, localization and multiple detection Objects in pictures or videos. Identifying objects in a video or image stream can be done through processes such as pre-processing. segmentation, foreground and background extraction, feature extraction. We will try to add new features related to the real world in future versions. In today's scenario, the Single Shot Multi-Box Detector (SSD) algorithm is considered the fastest approach for object detection using a single layer of a convolutional network. In our paper, we have focused on enhancing the classification accuracy of object detection while maintaining the algorithm's speed. Experimental results have confirmed that our proposed improved SSD algorithm achieves high accuracy.

Enhancing Strawberry Harvesting Efficiency through Yolo-v7 Object Detection Assessment

Strawberry fruits which are rich in vitamin A and carotenoids offer benefits for maintaining healthy epithelial tissues and promoting maturity and growth. The intensive cultivation and swift maturation of strawberries make them susceptible to premature harvesting, leading to spoilage and financial losses for farmers. This underscores the need for an automated detection method to monitor strawberry development and accurately identify growth phases of fruits. To address this challenge, a dataset called Strawberry-DS, comprising 247 images captured in a greenhouse at the Agricultural Research Center in Giza, Egypt, is utilized in this research. The images of the dataset encompass various viewpoints, including top and angled perspectives, and illustrate six distinct growth phases: "green", “red”, "white", "turning", "early-turning" and "late-turning". This study employs the Yolo-v7 approach for object detection, enabling the recognition and classification of strawberries in different growth phases. The achieved mAP@.5 values for the growth phases are as follows: 0.37 for "green," 0.335 for "white," 0.505 for "early-turning," 1.0 for "turning," 0.337 for "late-turning," and 0.804 for "red". The comprehensive performance outcomes across all classes are as follows: precision at 0.792, recall at 0.575, mAP@.5 at 0.558, and mAP@.5:.95 at 0.46. Notably, these results show the efficacy of the proposed research, both in terms of performance evaluation and visual assessment, even when dealing with distracting scenarios involving imbalanced label distributions and unclear labeling of developmental phases of the fruits. This research article yields advantages such as achieving reasonable and reliable identification of strawberries, even when operating in real-time scenarios which also leads to a decrease in expenses associated with human labor.

3D detection transformer: Set prediction of objects using point clouds

Object detection in 3D scenes rely on two main methods: detection based on proposals (two-stage detectors) or detections based on anchors (single-stage detectors), similar to approaches for object detection in 2D. In this paper, we propose the 3DeTR framework that produces 3D detections without the use of anchors or proposals, allowing training of the entire neural network in an end-to-end manner. Raw point cloud scenes are augmented and input into distance-and-reflectiveness-based feature extractor to produce representative points. Then, a transformer encoder–decoder module learns the local object relations and global context to generate parallel detections, which are then passed to a set-based loss function to map predictions to the set of ground truth labels uniquely. The model’s architecture produces 3D detections by regressing directly with the set of ground truths without the need for anchors or proposals, which are bottlenecks for object detection performances. We tested the framework on the KITTI Vision Benchmark Suite 3D object detection dataset, achieving results on par with the state-of-the-art: 80.37 AP on Cars (Moderate) class and 47.92 AP on Pedestrians (Moderate) class.

Review of Machine-Learning Approaches for Object and Component Detection in Space Electro-optical Satellites

Review of Machine-Learning Approaches for Object and Component Detection in Space Electro-optical Satellites

SEGANet: 3D object detection with shape-enhancement and geometry-aware network

3D object detection approaches from point clouds develop rapidly. However, the distribution of point clouds is unbalanced in the real scene, and thus the distant or occluded objects suffer from too few points to be perceived. This case damages the overall detection accuracy. Hence, we propose a novel two-stage 3D object detection framework, the Shape-Enhancement and Geometry-Aware Network (SEGANet), which aims to mitigate the negative impact of unbalanced point distribution for boosting detection performance. In stage 1, we first capture fine-grained structural knowledge with the assistance of point-wise features from voxels to generate proposals. And in stage 2, we construct a shape enhancement module to reconstruct complete surface points for objects within proposals, then establish an elaborate geometric relevance-aware Transformer module to aggregate high-correlated feature pairs of reconstructed-known parts and decode vital geometric relations of aggregated features. Thus, critical geometric clues are supplied for objects from the data and feature levels, achieving enhanced features for box refinement. Extensive experiments on KITTI and Waymo datasets show that SEGANet achieves low model complexity and excellent detection accuracy, especially leading the baseline method by 2.18% gain in overall detection accuracy and 1.8% gain in average accuracy of weakly sensing objects. This verifies that SEGANet effectively alleviates the impact of point imbalance to significantly boost detection performance.