Small Objects Research Articles

SMALL OBJECT DETECTION APPROACH BASED ON ENHANCED SINGLE-SHOT DETECTOR FOR DETECTION AND RECOGNITION OF INDONESIAN TRAFFIC SIGNS

The detection and recognition of traffic signs are crucial components of advanced driving assistance systems (ADAS) that enhance road safety. Current traffic sign detection and recognition model technology is proficient in identifying and interpreting traffic signs. However, for accurate detection and recognition, the traffic sign in the image must be of a certain minimum pixel size or distance from the driver's sight line for proper detection. The ADAS system should be capable of detecting and recognizing road traffic signs from a considerable distance as they come into the driver's line of vision. The higher the vehicle speed, the greater the distance required for the sign to be detected and recognized, allowing the driver sufficient time to react according to the sign's meaning. Addressing these challenges, this research proposes an enhanced version of the single shot detector (SSD) algorithm, commonly used in object detection, to improve the algorithm's ability to detect small objects. The proposed method involves adding an auxiliary layer module to the original SSD architecture to increase the feature map resolution and expand the conventional layer's receptive space. With the Enhanced SSD algorithm, the detection capability of the SSD can be significantly enhanced in terms of accuracy. The limitations of this study are related to the influence of occlusion and clutter, which might affect the performance of object detection, especially for small objects, which are more susceptible to being influenced by various factors. The research results demonstrate that Enhanced SSD improves object detection accuracy compared to the original SSD, with a mean average precision (mAP) of 97.87 compared to 95.35 for detecting 21 traffic signs in Indonesia.

Improving rail transit security with enhanced YOLOv5 obstacle detection

Abstract As the economy advances and rail transit technology progresses, rail transit plays an important role in alleviating urban traffic pressure and promoting regional economic development. However, railway safety faces potential hazards such as foreign object intrusion in the trackside environment, seriously threatening the safety of trains and passengers. Therefore, to achieve high-performance recognition of obstacle target images in rail transit, a small sample object detection algorithm YOLOv5-RTO, which using EVC (Explicit Visual Center) attention mechanism and lightweight up-sampling operator CARAFE (Content-Aware ReAssembly of FEatures) to improve YOLOv5, has been proposed. Based on the self-collected image data of rail transit obstacles, a dense and time-varying dataset was created within the rail transit system, and data augmentation methods were used to reduce the impact of small sample datasets on the detection performance of subsequent algorithm. Using this dataset, target detection tests were conducted on rail transit obstacle images to analyze the actual performance of the training model. The experiment results indicate that the enhanced YOLOv5-RTO algorithm integrates the advantages of EVC and CARAFE while preserving the lightweight and accurate characteristics of YOLOv5. The improved algorithm scheme has an average inference time of 1.3 ms per frame in the field of rail transit images, achieving an overall mAP of 96.5%, an increase of 1.8% compared to the original algorithm. It has a certain engineering application value and provides reference for further optimizing the safety system of urban rail transit.

Rethinking the Non-Maximum Suppression Step in 3D Object Detection from a Bird’s-Eye View

In camera-based bird’s-eye view (BEV) 3D object detection, non-maximum suppression (NMS) plays a crucial role. However, traditional NMS methods become ineffective in BEV scenarios where the predicted bounding boxes of small object instances often have no overlapping areas. To address this issue, this paper proposes a BEV intersection over union (IoU) computation method based on relative position and absolute spatial information, referred to as B-IoU. Additionally, a BEV circular search method, called B-Grouping, is introduced to handle prediction boxes of varying scales. Utilizing these two methods, a novel NMS strategy called BEV-NMS is developed to handle the complex prediction boxes in BEV perspectives. This BEV-NMS strategy is implemented in several existing algorithms. Based on the results from the nuScenes validation set, there was an average increase of 7.9% in mAP when compared to the strategy without NMS. The NDS also showed an average increase of 7.9% under the same comparison. Furthermore, compared to the Scale-NMS strategy, the mAP increased by an average of 3.4%, and the NDS saw an average improvement of 3.1%.

Enhanced Abandoned Object Detection through Adaptive Dual-Background Modeling and SAO-YOLO Integration

Abandoned object detection is a critical task in the field of public safety. However, existing methods perform poorly when detecting small and occluded objects, leading to high false detection and missed detection rates. To address this issue, this paper proposes an abandoned object detection method that integrates an adaptive dual-background model with SAO-YOLO (Small Abandoned Object YOLO). The goal is to reduce false and missed detection rates for small and occluded objects, thereby improving overall detection accuracy. First, the paper introduces an adaptive dual-background model that adjusts according to scene changes, reducing noise interference in the background model. When combined with an improved PFSM (Pixel-based Finite State Machine) model, this enhances detection accuracy and robustness. Next, a network model called SAO-YOLO is designed. Key improvements within this model include the SAO-FPN (Small Abandoned Object FPN) feature extraction network, which fully extracts features of small objects, and a lightweight decoupled head, SODHead (Small Object Detection Head), which precisely extracts local features and enhances detection accuracy through multi-scale feature fusion. Finally, experimental results show that SAO-YOLO increases mAP@0.5 and mAP@0.5:0.95 by 9.0% and 5.1%, respectively, over the baseline model. It outperforms other advanced detection models. Ultimately, after a series of experiments on the ABODA, PETS2006, and AVSS2007 datasets, the proposed method achieved an average detection precious of 91.1%, surpassing other advanced methods. It significantly outperforms other advanced detection methods. This approach notably reduces false and missed detections, especially for small and occluded objects.

Vehicle and Pedestrian Detection Based on Improved YOLOv7-Tiny

To improve the detection accuracy of vehicles and pedestrians in traffic scenes using object detection algorithms, this paper presents modifications, compression, and deployment of the single-stage typical algorithm YOLOv7-tiny. In the model improvement section: firstly, to address the problem of small object missed detection, shallower feature layer information is incorporated into the original feature fusion branch, forming a four-scale detection head; secondly, a Multi-Stage Feature Fusion (MSFF) module is proposed to fully integrate shallow, middle, and deep feature information to extract more comprehensive small object information. In the model compression section: the Layer-Adaptive Magnitude-based Pruning (LAMP) algorithm and the Torch-Pruning library are combined, setting different pruning rates for the improved model. In the model deployment section: the V7-tiny-P2-MSFF model, pruned by 45% using LAMP, is deployed on the embedded platform NVIDIA Jetson AGX Xavier. Experimental results show that the improved and pruned model achieves a 12.3% increase in mAP@0.5 compared to the original model, with parameter volume, computation volume, and model size reduced by 76.74%, 7.57%, and 70.94%, respectively. Moreover, the inference speed of a single image for the pruned and quantized model deployed on Xavier is 9.5 ms.

AF-DETR: efficient UAV small object detector via Assemble-and-Fusion mechanism

AF-DETR: efficient UAV small object detector via Assemble-and-Fusion mechanism

Self-supervised monocular depth estimation based on improved dense network and wavelet decomposition

Traditional self-supervised monocular depth estimation models do not adequately extract and fuse shallow features, which can easily lead to problems such as missed detection of small objects and blurred object edges. To address the above problems, this paper proposes a self-supervised monocular depth estimation model based on improved dense network and wavelet decomposition. The model follows the structure of U-net, in which the encoder adopts an improved dense connection method to improve the feature extraction and fusion capabilities of the encoder; at the same time, a detail enhancement module is added to the jump connection to further refine and integrate the multi-scale features output by the encoder; finally, the wavelet decomposition method is introduced in the decoder, forcing the decoder to pay more attention to high-frequency information and achieve refined processing of image edges. Experimental results show that the depth estimation model proposed in this paper has a stronger ability to capture small object features, and the edges of the generated depth map are clearer and more accurate.

A Novel YOLOv5_ES based on lightweight small object detection head for PCB surface defect detection

In the manufacturing process of printed circuit boards (PCBs), surface defects have a significant negative impact on product quality. Considering that traditional object detection algorithms have low accuracy in handling PCB images with complex backgrounds, various types, and small-sized defects, this paper proposes a PCB defect detection algorithm based on a novel YOLOv5 multi-scale attention mechanism(EMA) spatial pyramid dilated Convolution (SPD-Conv) (YOLOv5_ES) network improved YOLOv5s framework. Firstly, the detection head is optimized by removing medium and large detection layers, fully leveraging the capability of the small detection head to identify minor target defects. This approach not only improves model accuracy but also achieves lightweighting. Secondly, in order to further reduce the number of parameters and computational costs, the SPD-Conv is introduced to improve the feature extraction capability by reducing information loss. Thirdly, a EMA module is introduced to fuse context information of different scales, enhancing the model’s generalization ability. Compared to the YOLOv5s model, there is a 3.1% improvement in mean average precision (mAP0.5), a 55.8% reduction in model parameters, and a 4.8% reduction in giga floating-point operations per second (GFLOPs). These results demonstrate a significant improvement in both accuracy and model parameter efficiency.

A small object detection architecture with concatenated detection heads and multi-head mixed self-attention mechanism

A small object detection architecture with concatenated detection heads and multi-head mixed self-attention mechanism

A Novel Pre-Processing Approach and Benchmarking Analysis for Faster, Robust, and Improved Small Object Detection Methods

A Novel Pre-Processing Approach and Benchmarking Analysis for Faster, Robust, and Improved Small Object Detection Methods

Innovative methods for trackability and identification improvement of small objects for space traffic management

Technology solutions and operational implications need a evaluation for their crucial role in tracking and identification methodologies. The growth of objects in Earth orbit have made it more important that every space system can be accounted for to be able manage the collision risk between all space objects. Tracking and identification are intimately related as enhanced tracking capabilities minimizes the need for special identification techniques. For both features, the difference between active and passive systems are discussed. Tracking is primarily optical or radar; which of these that is being used drives the most operationally-relevant identification solutions. For example, radar tags can be made to be respond to the radio frequency (RF) energy from the radar and laser reflectors can be completely passive. Whereas identification aids for use by optical tracking are active light sources (of some selected frequency) requiring a power source.

Worldtube excision method for intermediate-mass-ratio inspirals: Self-consistent evolution in a scalar-charge model

This is a third installment in a program to develop a method for alleviating the scale disparity in binary black hole simulations with mass ratios in the intermediate astrophysical range, where simulation cost is prohibitive while purely perturbative methods may not be adequate. The method is based on excising a “worldtube” around the smaller object, much larger than the object itself, replacing it with an analytical model that approximates a tidally deformed black hole. Previously [N. A. Wittek , Worldtube excision method for intermediate-mass-ratio inspirals: Scalar-field model in 3+1 dimensions, ] we have tested the idea in a toy model of a scalar charge in a fixed circular geodesic orbit around a Schwarzschild black hole, solving for the massless Klein-Gordon field in 3+1 dimensions on the p platform. Here we take the significant further step of allowing the orbit to evolve radiatively, in a self-consistent manner, under the effect of back-reaction from the scalar field. We compute the inspiral orbit and the emitted scalar-field waveform, showing a good agreement with perturbative calculations in the adiabatic approximation. We also demonstrate how our simulations accurately resolve postadiabatic effects (for which we do not have perturbative results). In this work we focus on quasicircular inspirals. Our implementation is publicly accessible in the p numerical relativity code. Published by the American Physical Society 2024

SOD‐YOLO: Small Object Detection Network for UAV Aerial Images

With the rapid development of the UAV industry, object detection using UAV has become a research hotspot. However, most current object detection models based on deep learning have large parameter counts and are difficult to deploy on embedded devices with limited memory and computational power. To address this problem, a Small Object Detection network for UAV aerial images SOD‐YOLO based on YOLOv8 is proposed, which can meet the application requirements of resource‐constrained devices while ensuring the detection accuracy. First, cross‐domain fusion attention (CDFA) mechanism is proposed to build the C2f‐Attention module in this paper, which is embedded in the backbone network in order to improve the extraction capability of key object features. Meanwhile, the AIFI_LSPE feature fusion module with improved RT‐DETR and the IoU‐aware query selection mechanism are added to the path aggregation network to improve the accuracy of multi‐scale object detection. In addition, in order to balance the sample size ratio and improve the robustness of the network model, we make a new UAV image dataset named VisDrone2019 Extended Edition (VDEE) using images from the VisDrone2019 and UAVDT public datasets. Finally, Shape‐IoU is used as a loss function to reduce the difference between the object GT frame and the detection frame. Experiments show that SOD‐YOLO has a mAP@0.5 of 42.8% in the VDEE dataset, which is increased by 5.1% over YOLOv8. In the VisDrone2019 dataset mAP@0.5 is 39.2%, an improvement of 5.8% over YOLOv8. © 2024 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.

Excellent tomato detector based on pruning and distillation to balance accuracy and lightweight

Rapid and accurate detection of tomatoes in natural plantation environments is the linchpin for smart harvesting. However, the complexity of existing models necessitates high hardware computing power, constraining the deployment of tomato recognition models in picking robots. In this study, the overall scheme framework of the vision algorithm is divided into three stages. In the first stage, based on YOLOv8s, a small object detection layer was added along with hyperparameter optimization, resulting in the proposed YOLOv8-TomatoTeacher (YOLOv8-TT) to improve the accuracy and precise positioning of tomato detection, laying the foundation for subsequent operations. In the second stage, the pruning algorithm is applied to the trained YOLOv8-TT model to trim and optimize redundant channels, yielding the pruned lightweight YOLOv8-TomatoStudent (YOLOv8-TS) model. In the third stage, using YOLOv8-TT as the teacher model and YOLOv8-TS as the student model, the detection accuracy of YOLOv8-TS was further enhanced through knowledge distillation, thus obtaining the YOLOv8-TomatoDetector (YOLOv8-TD) that considers both detection speed and accuracy. Experimental results indicate that YOLOv8-TD demonstrates significant advantages in parameter quantity and model complexity, with only 0.35 M Params and 4.4G GFlops, which are respectively reduced by 96.86 % and 84.51 % compared with baseline model YOLOv8s. In addition, YOLOv8-TD achieved a 1.4 % increase in mAP, and the detection speed of 354f/s satisfies real-time requirements. The model proposed in this study can swiftly and accurately detect dense small tomato targets, reducing the deployment cost and difficulty of the robot vision system.

Improved Small Object Detection Algorithm CRL-YOLOv5.

Detecting small objects in images poses significant challenges due to their limited pixel representation and the difficulty in extracting sufficient features, often leading to missed or false detections. To address these challenges and enhance detection accuracy, this paper presents an improved small object detection algorithm, CRL-YOLOv5. The proposed approach integrates the Convolutional Block Attention Module (CBAM) attention mechanism into the C3 module of the backbone network, which enhances the localization accuracy of small objects. Additionally, the Receptive Field Block (RFB) module is introduced to expand the model's receptive field, thereby fully leveraging contextual information. Furthermore, the network architecture is restructured to include an additional detection layer specifically for small objects, allowing for deeper feature extraction from shallow layers. When tested on the VisDrone2019 small object dataset, CRL-YOLOv5 achieved an mAP50 of 39.2%, representing a 5.4% improvement over the original YOLOv5, effectively boosting the detection precision for small objects in images.

Exploring the application of a Bayesian framework in transfer path analysis

Transfer path analysis (TPA) is a methodology widely used in transmission identification of vibration and noise. Among the TPA families, the two prevailing approaches are classical TPA with the matrix inversion method and operational transfer path analysis (OTPA). Both methods rely on measurements within costly and complicated overdetermined systems. These experiment-based methods are also limited by sufficient space for the sensor mounting, which can result in difficulty when analyzing the transfer path in small objects. In this paper, we use an experimental setup that mimics an engine mounted on a suspension with resilient connections. The severity of errors stemming from the application of underdetermined measurement systems is assessed, alongside an exploration of how the Bayesian framework can be integrated into the analysis to mitigate errors and enhance the performance of TPA.

YOLOv7-KDT: An ensemble model for pomelo counting in complex environment

Pomelo counting recognition in complex environments faces many difficulties, mainly including changes in illumination, occlusion, and complex backgrounds. These factors contribute to the poor recognition performance of early models such as Faster R-CNN, SSD, and Efficientdet. This study optimized the YOLOv7 object detection network through several strategies for this issue. The main optimization strategies are: (1) Introducing the SEAM module to improve the detection ability for occluded objects; (2) Using the AIFI attention mechanism to enhance the network’s recognition of small objects; (3) Adopting the LAMP pruning technique to reduce model complexity; (4) Utilizing multi-teacher knowledge distillation for model compression and performance enhancement. The results show that under the assisted learning of the student module by the multi-teacher module distillation, the proposed method improves mAP by more than 1.9% compared to the aforementioned early models and other networks. After LAMP pruning, the model size is reduced by at least 93.6%. After knowledge distillation processing, its mAP is increased from 84.2% to 91.1%, achieving high-precision pomelo counting recognition. This study effectively solves automatic pomelo recognition and counting in complex environments. The method can be generalized to other agricultural product detection and statistics scenarios, having broad application prospects.

Modelling-assisted geometrical optimization of colloidal quantum color convertor based pixels fabricated by dielectrophoretic directed assembly

Hypothesis: Building competitive color conversion pixels for microdisplays made of semiconductor nanocrystals requires reaching a deposition thickness high enough to absorb all the blue light from the backlight unit. In the case of dielectrophoretic directed assembly of such nanocrystals, modeling and simulations may help understand what the intrinsic limitations of the process are, and may be used to propose new assembly routes.Experiments: A theoretical model of dielectrophoretic interactions between polarizable nano-spheres and an electrostatically patterned substrate has been developed. Monte Carlo simulations have been run using this model to rationalize the effects of parameters driving the dielectrophoretic directed assembly and to find optimal deposition conditions for reaching a maximal thickness of nanocrystal pixels. Experiments with CdSe quantum plates and with alumina spheres embedding quantum plates (micro-pearls) have been carried out and compared to the model.Findings: Modeling and simulations reveal that the directed assembly of semiconductor nanocrystals is limited essentially by the small object size, which sets the maximum dielectrophoretic force they can undergo. They indicate that using larger objects should allow reaching unprecedented assembly heights, but will induce lateral extension of the assembly. This trade-off has been illustrated with diagrams in the parameter space and confirmed experimentally with micro-pearls.

AUTOMATIC PANEL QUALITY MONITORING SYSTEM USING VISION-BASED ABNORMAL OBJECT DETECTION

The rate of “Not Good” products due to the presence of abnormal objects during the quality control inspection of mobile phone’s panels is as high as 38 percent without supervision and 19 percent with naked eye’s inspection. To reduce the loss of small abnormal objects and improve the recognition rate of object detection of automatic panel quality monitoring, the paper proposes abnormal object recognition based on a template matching and subtract background technique. Firstly, the object recognition network based on convolutional neural networks (CNNs) is introduced initially. Then, the Canny Edge Detection method improves the image input quality. Next, template matching is utilized to discover things that depart from a standard pattern or a predetermined model. Models for background reduction will be implemented instantly into smartphone panel quality checking systems. Continually, a vision-based quality control (QC) strategy will be developed. Lastly, experimental results from a practical vision-based automatic anomalous object identification system demonstrate the viability and enhancement of the suggested strategy.

Text-Guided Multi-Class Multi-Object Tracking for Fine-Grained Maritime Rescue

The rapid development of remote sensing technology has provided new sources of data for marine rescue and has made it possible to find and track survivors. Due to the requirement of tracking multiple survivors at the same time, multi-object tracking (MOT) has become the key subtask of marine rescue. However, there exists a significant gap between fine-grained objects in realistic marine rescue remote sensing data and the fine-grained object tracking capability of existing MOT technologies, which mainly focuses on coarse-grained object scenarios and fails to track fine-grained instances. Such a gap limits the practical application of MOT in realistic marine rescue remote sensing data, especially when rescue forces are limited. Given the promising fine-grained classification performance of recent text-guided methods, we delve into leveraging labels and attributes to narrow the gap between MOT and fine-grained maritime rescue. We propose a text-guided multi-class multi-object tracking (TG-MCMOT) method. To handle the problem raised by fine-grained classes, we design a multi-modal encoder by aligning external textual information with visual inputs. We use decoding information at different levels, simultaneously predicting the category, location, and identity embedding features of objects. Meanwhile, to improve the performance of small object detection, we also develop a data augmentation pipeline to generate pseudo-near-infrared images based on RGB images. Extensive experiments demonstrate that our TG-MCMOT not only performs well on typical metrics in the maritime rescue task (SeaDronesSee dataset), but it also effectively tracks open-set categories on the BURST dataset. Specifically, on the SeaDronesSee dataset, the Higher Order Tracking Accuracy (HOTA) reached a score of 58.8, and on the BURST test dataset, the HOTA score for the unknown class improved by 16.07 points.