Non-maximum Suppression Algorithm Research Articles

Defect intelligent recognition of membrane product based on deep learning

Defect detection plays a crucial role in the manufacturing industry, ensuring the quality of industrial products. Despite advancements in this field, current defect detection methods face two primary challenges: (1) extracting visually similar features from the background poses difficult and (2) these methods struggle to identify tiny defects in the target objects. We present a feature enhancement module called the SE-CAR, which aims to handle the identified problems effectively. This module is designed to efficiently capture tiny defects in images, prioritize defect information features, and ultimately enhance the model’s predictive performance and accuracy in defect recognition tasks. In addition, Distance-IoU Non-Maximum Suppression is employed as a substitution for the original Non-Maximum Suppression. This enhances the recognition accuracy of bounding boxes, ensuring that the model maintains high detection accuracy even after complex scenarios. Moreover, the proposed methodology exhibits broad applicability across a wide spectrum of prevalent defect detection paradigms. In empirical experiments, we employ the YOLOv7 architecture as the foundation framework, integrating the proposed methodologies for the purpose of detecting defects in the membrane. The empirical evidence demonstrates a notable improvement in the performance of detecting defects in membrane products using the SE-CAR feature enhancement module and Distance-IoU Non-Maximum Suppression algorithm. In comparison to baseline networks, an increase in mAp by 6.29%, precision by 1.63%, and recall by 8.76%.

Research and Experiment on Miss-Seeding Detection of Potato Planter Based on Improved YOLOv5s

In order to improve the performance of potato planter, reduce miss-seeding rates, enhance the overall quality of the seeding operation, and ultimately increase the yield of the potato, it is necessary to implement effective technical means to monitor and identify the miss-seeding issues during the seeding process. The existing miss-seeding detection technologies commonly use sensors to monitor, but such technologies are easily affected by factors like heavy dust and strong vibrations, resulting in poor interference resistance and adaptability. Therefore, this study aims to explore and apply deep learning algorithms to achieve real-time monitoring of the miss-seeding phenomenon in potato planter during the planting process. Considering both the lightweight of the miss-seeding detection model and its practical deployment, this study selects and adapts the YOLOv5s algorithm to achieve this goal. Firstly, the attention mechanism is integrated into the backbone network to suppress background interference and improve detection accuracy. Secondly, the non-maximum suppression algorithm is improved by replacing the original IoU-NMS with the Soft-NMS algorithm to enhance the bounding box regression rate and reduce missed detections of potato seeds due to background overlap or occlusion. Experimental results show that the accuracy of the improved algorithm in detecting miss-seeding increased from 96.02% to 98.30%, the recall rate increased from 96.31% to 99.40%, and the mean average precision (mAP) improved from 99.12% to 99.40%. The improved model reduces missed and false detections, provides more precise target localization, and is suitable for miss-seeding detection in natural environments for potato planter, providing technical and theoretical support for subsequent intelligent reseeding in potato planter.

Enhancing detection accuracy of highly overlapping targets in agricultural imagery using IoA-SoftNMS algorithm across diverse image sizes

Object detection is crucial for agricultural monitoring and agriculture automation, yet conventional non-maximum suppression (NMS) algorithms often struggle in agricultural scenes with complex backgrounds and highly overlapping targets, especially in intersection over union (IoU) threshold setting and abnormal non-maximum removal. This study proposed the soft non-maximum suppression based on intersection over area (IoA-SoftNMS) algorithm, an advanced NMS algorithm designed to enhance detection accuracy in such challenging agricultural environments. IoA-SoftNMS employs intersection over area (IoA) to quantify overlap and integrate SoftNMS to boost overall detection precision and effectiveness. Validation experiments demonstrated that IoA-SoftNMS significantly increased detection accuracy, improving mean average precision (mAP) by 0.7 % to 3.5 % across various object detection models, without necessitating model retraining. Importantly, these enhancements barely affect the detection time. IoA-SoftNMS exhibited robust adaptability in detecting large images, especially when integrated with the slicing-aided hyper inference (SAHI) technique, where it notably improved detection accuracy during the slicing process. Application of IoA-SoftNMS elevates detection accuracy in medium (1000 × 3300 pixels) and large (2500 × 3400 pixels) images from 93 % to 96.9 % and 90.8 % to 95.3 %, respectively. These findings underscored the effectiveness of the proposed IoA-SoftNMS algorithm in improving object detection accuracy in agricultural settings and its potential to alleviate challenges posed by large image sizes, offering avenues for enhancing agricultural production precision and intelligence.

Bridging the gap between object detection in close-up and high-resolution wide shots

Recent years have seen a significant rise in gigapixel-level image/video capture systems and benchmarks with high-resolution wide (HRW) shots. Different from close-up shots like MS COCO, the higher resolution and wider field of view raise new research and application problems, such as how to perform accurate and efficient object detection with such large input in low-power edge devices like UAVs. There are several unique challenges in HRW shots. (1) Sparse information: the objects of interest cover less area. (2) Various scale: there is 10 to 100× object scale change in one single image. (3) Incomplete objects: the sliding window strategy to handle the large input leads to truncated objects at the window edge. (4) Multi-scale information: it is unclear how to use multi-scale information in training and inference. Consequently, directly using a close-up detector leads to inaccuracy and inefficiency. In this paper, we systematically investigate this problem and bridge the gap between object detection in close-up and HRW shots, by introducing a novel sparse architecture that can be integrated with common networks like ConvNet and Transformer. It leverages alternative sparse learning to complementarily fuse coarse-grained and fine-grained features to (1) adaptively extract valuable information from (2) different object scales. We also propose a novel Cross-window Non-Maximum Suppression (C-NMS) algorithm to (3) improve the box merge from different windows. Furthermore, we propose a (4) simple yet effective multi-scale training and inference strategy to improve accuracy. Experiments on two benchmarks with HRW shots, PANDA and DOTA-v1.0, demonstrate that our methods significantly improve accuracy (up to 5.8%) and speed (up to 3×) over SotAs, for both ConvNet or Transformer based detectors, on edge devices. Our code is open-sourced and available at https://github.com/liwenxi/SparseFormer.

An Efficient and Accurate Quality Inspection Model for Steel Scraps Based on Dense Small-Target Detection

Scrap steel serves as the primary alternative raw material to iron ore, exerting a significant impact on production costs for steel enterprises. With the annual growth in scrap resources, concerns regarding traditional manual inspection methods, including issues of fairness and safety, gain increasing prominence. Enhancing scrap inspection processes through digital technology is imperative. In response to these concerns, we developed CNIL-Net, a scrap-quality inspection network model based on object detection, and trained and validated it using images obtained during the scrap inspection process. Initially, we deployed a multi-camera integrated system at a steel plant for acquiring scrap images of diverse types, which were subsequently annotated and employed for constructing an enhanced scrap dataset. Then, we enhanced the YOLOv5 model to improve the detection of small-target scraps in inspection scenarios. This was achieved by adding a small-object detection layer (P2) and streamlining the model through the removal of detection layer P5, resulting in the development of a novel three-layer detection network structure termed the Improved Layer (IL) model. A Coordinate Attention mechanism was incorporated into the network to dynamically learn feature weights from various positions, thereby improving the discernment of scrap features. Substituting the traditional non-maximum suppression algorithm (NMS) with Soft-NMS enhanced detection accuracy in dense and overlapping scrap scenarios, thereby mitigating instances of missed detections. Finally, the model underwent training and validation utilizing the augmented dataset of scraps. Throughout this phase, assessments encompassed metrics like mAP, number of network layers, parameters, and inference duration. Experimental findings illustrate that the developed CNIL-Net scrap-quality inspection network model boosted the average precision across all categories from 88.8% to 96.5%. Compared to manual inspection, it demonstrates notable advantages in accuracy and detection speed, rendering it well suited for real-world deployment and addressing issues in scrap inspection like real-time processing and fairness.

CCBA-NMS-YD: A Vehicle Pedestrian Detection and Tracking Method Based on Improved YOLOv7 and DeepSort

In this paper, we propose a vehicle pedestrian detection and tracking method based on the improved YOLOv7 and DeepSort algorithms. We aim to improve the quality of vehicle pedestrian detection and tracking, addressing the challenges that current commercially available autonomous driving technologies face in complex and changing road traffic situations. First, the NMS (non-maximum suppression) algorithm in YOLOv7 is replaced with a modified Soft-NMS algorithm to ensure that targets can be accurately detected at high densities, and second, the CCBA (coordinate channel attention module) attention mechanism is incorporated to improve the feature extraction and perception capabilities of the network. Finally, a multi-scale feature network is introduced to extract features of small targets more accurately. Finally, the MobileNetV3 lightweight module is introduced into the feature extraction network of DeepSort, which not only reduces the number of model parameters and network complexity, but also improves the tracking performance of the target. The experimental results show that the improved YOLOv7 algorithm improves the average detection accuracy by 3.77% compared to that of the original algorithm; on the MOT20 dataset, the refined DeepSort model achieves a 1.6% increase in MOTA and a 1.9% improvement in MOTP; in addition, the model volume is one-eighth of the original algorithm. In summary, our model is able to achieve the desired real-time and accuracy, which is more suitable for autonomous driving.

ODCS-YOLO detection algorithm for rail surface defects based on omni-dimensional dynamic convolution and context augmentation module

Abstract To solve the problems of easy miss and false detection on rail surface defects caused by small size, dense target, and high similarity between features and background, this paper proposed an improved detection algorithm in complex background. First, the conventional convolution of YOLOv5 backbone network is replaced with omni-dimensional dynamic convolution (ODConv), which improves the feature extraction capability of the network without increasing the computational cost; second, to improve the model's performance in detecting tiny objects, a two-layer context augmentation module(CAM) is introduced into the path aggregation network(PAN) structure; finally, the traditional non-maximum suppression(NMS) algorithm is replaced by the Soft-NMS algorithm in the network post-processing to reduce the false-alarm and miss-rate. The experimental results on the Railway Track Fault Detection public dataset show that the OD-YOLO (OD stands for ODConv) and C-PAN(CAM module is introduced into PAN) structures could achieve better performance in the same type of improved algorithms; compared with the baseline algorithm YOLOv5, the ODCS-YOLO (OD stands for ODConv, C stands for CAM, and S stands for Soft-NMS) algorithm improves the precision by 12.4%, the recall by 3.6%, the map50 by 8.6% and the GFLOPs is reduced by 0.6. Compared with seven classical object detection algorithms, the ODCS-YOLO algorithm achieves the highest detection accuracy, which makes it able to meet the real-time detection requirements of rail surface defects in real working conditions. The ODCS-YOLO model provides certain technical support for the defects detection and a new method for the detection of dense small objects.

The effect of pre- and post-processing techniques on tree detection in young forest stands from images of snow cover using YOLO neural networks

A neural network model for individual tree detection was developed based on the YOLOv4 architecture, which underwent additional preprocessing and postprocessing steps. The preprocessing step involved expanding the dataset by randomly cutting fragments from images, calculating anchor frame sizes using the K-means clustering algorithm, and discarding anchor frames that were too small a priori. The existing post-processing block of the YOLO architecture was modified by giving more weight to false positives in the error function and using the non-maximum suppression algorithm. Baseline neural networks from the YOLOv4 and YOLOv5 architectures, each in two versions (pre-trained and not pre-trained on the MS COCO dataset), were used for comparison without any additional modifications. In the overgrown experimental field, multi-season aerial copter surveys and ground counts were conducted on several sample plots to gather data. Comparison of multi-season aerial photographs with ground-count data showed that the best images in terms of the percentage of visually identifiable trees were those taken during the snowy season and when there was no foliage. Using these images and some additional images, we manually created a dataset on which we trained and tested neural network models. The model we developed showed significantly better results (2 to 10 times better) on the mAP 0.5 metric compared to the alternatives we considered.

Uncertainty model-based edge detection technology in badminton

As virtual reality technology develops, the analysis and processing of video content have become hot spots in the field of computer vision. Video Action Detection aims to locate features in network video, and its research spans many fields, such as computer vision and spatial prediction. In view of the problem of low-efficiency classification models and inaccurate localization of small-scale targets in complex scenes, we propose a novel method to generate candidate intervals for action detection. The action recognition model is adopted to generate the action score sequence on the video time series. We also propose the uncertainty model of the descending pose detection algorithm. The pre-reaction phase generates a candidate list in the form of concatenated videos containing exactly the same pose to detect action poses that are not identical and of non-maximum duration. Experiments with traditional target detection and multiple deep learning models show that the proposed Non-Maximum Suppression algorithm has a strong ability to extract neural network features. Furthermore, compared with traditional ATSS and Faster R-CNN methods, the detection quality and performance are improved by more than 15.2% and 7.8%, respectively. Our method can fully utilize perception information to improve the quality of decision planning and plays a connecting role between perception fusion and decision planning.

Intelligent Monitoring of Protection Devices in Power system with Enhanced Faster R-CNN

Relay protection devices are necessary to guard the power system safety and stability. With the significant number of substations and relay protection devices, the maintenance workload has become difficult due to limited manpower, posing hidden dangers to the reliable operation of protection devices. In view of this, this article proposes an enhanced Faster R-CNN algorithm to diagnose the relay protection devices based on image monitoring. The proposed algorithm uses RestNet50 as the main tool to recognize features from input images to generate feature maps. Meanwhile, to improve the image detection accuracy, the Non-Maximum Sup-pression (NMS) algorithm in regional suggestion network (RPN) is optimized from solid thresh-old to soft threshold. By combining the images captured by the surveillance camera, intelligent inspections are conducted on the status of the pressing plate, fiber bending, external object retention, and other items in the relay protection device. A great number of samples are used to train networks for different tasks. The results indicate that the enhanced Faster R-CNN has a higher image recognition rate, and the recognition accuracy of each inspection item for the protection device is higher than 95%.

Performance Optimization and Application Research of YOLOv8 Model in Object Detection

This paper focuses on the performance optimization and application of the YOLOv8 model in object detection. Firstly, the model architecture and optimization strategies of YOLOv8 are outlined, including network structure, feature extraction, and fusion, demonstrating its advantages in detection accuracy, inference speed, and robustness. Then, the training and inference processes of YOLOv8 are discussed in depth, addressing the issue of imbalance between positive and negative samples, and optimizing the forward propagation and non-maximum suppression algorithm. Through experimental comparisons, YOLOv8 outperforms other mainstream object detection models in multiple metrics. Finally, summarize the contributions and limitations of YOLOv8, and look forward to its future research directions and application prospects. This paper provides valuable references for the optimization and application of the YOLOv8 algorithm, aiming to provide theoretical and practical support for the further development of object detection technology in the field of computer vision.

STNet: A Time-Frequency Analysis-Based Intrusion Detection Network for Distributed Optical Fiber Acoustic Sensing Systems.

Distributed optical fiber acoustic sensing (DAS) is promising for long-distance intrusion-anomaly detection tasks. However, realistic settings suffer from high-intensity interference noise, compromising the detection performance of DAS systems. To address this issue, we propose STNet, an intrusion detection network based on the Stockwell transform (S-transform), for DAS systems, considering the advantages of the S-transform in terms of noise resistance and ability to detect disturbances. Specifically, the signal detected by a DAS system is divided into space-time data matrices using a sliding window. Subsequently, the S-transform extracts the time-frequency features channel by channel. The extracted features are combined into a multi-channel time-frequency feature matrix and presented to STNet. Finally, a non-maximum suppression algorithm (NMS), suitable for locating intrusions, is used for the post-processing of the detection results. To evaluate the effectiveness of the proposed method, experiments were conducted using a realistic high-speed railway environment with high-intensity noise. The experimental results validated the satisfactory performance of the proposed method. Thus, the proposed method offers an effective solution for achieving high intrusion detection rates and low false alarm rates in complex environments.

Working toward Solving Safety Issues in Human–Robot Collaboration: A Case Study for Recognising Collisions Using Machine Learning Algorithms

The monitoring and early avoidance of collisions in a workspace shared by collaborative robots (cobots) and human operators is crucial for assessing the quality of operations and tasks completed within manufacturing. A gap in the research has been observed regarding effective methods to automatically assess the safety of such collaboration, so that employees can work alongside robots, with trust. The main goal of the study is to build a new method for recognising collisions in workspaces shared by the cobot and human operator. For the purposes of the research, a research unit was built with two UR10e cobots and seven series of subsequent of the operator activities, specifically: (1) entering the cobot’s workspace facing forward, (2) turning around in the cobot’s workspace and (3) crouching in the cobot’s workspace, taken as video recordings from three cameras, totalling 484 images, were analysed. This innovative method involves, firstly, isolating the objects using a Convolutional Neutral Network (CNN), namely the Region-Based CNN (YOLOv8 Tiny) for recognising the objects (stage 1). Next, the Non-Maximum Suppression (NMS) algorithm was used for filtering the objects isolated in previous stage, the k-means clustering method and Simple Online Real-Time Tracking (SORT) approach were used for separating and tracking cobots and human operators (stage 2) and the Convolutional Neutral Network (CNN) was used to predict possible collisions (stage 3). The method developed yields 90% accuracy in recognising the object and 96.4% accuracy in predicting collisions accuracy, respectively. The results achieved indicate that understanding human behaviour working with cobots is the new challenge for modern production in the Industry 4.0 and 5.0 concept.

Gender Identification of Chinese Mitten Crab Juveniles Based on Improved Faster R-CNN

The identification of gender in Chinese mitten crab juveniles is a critical prerequisite for the automatic classification of these crab juveniles. Aiming at the problem that crab juveniles are of different sizes and relatively small, with unclear male and female characteristics and complex background environment, an algorithm C-SwinFaster for identifying the gender of Chinese mitten crab juveniles based on improved Faster R-CNN was proposed. This algorithm introduces Swin Transformer as the backbone network and an improved Path Aggregation Feature Pyramid Network (PAFPN) in the neck to obtain multi-scale high-level semantic feature maps, thereby improving the gender recognition accuracy of Chinese mitten crab male and female juveniles. Then, a self-attention mechanism is introduced into the region of interest pooling network (ROI Pooling) to enhance the model’s attention to the classification features of male and female crab juveniles and reduce background interference on the detection results. Additionally, we introduce an improved non-maximum suppression algorithm, termed Softer-NMS. This algorithm refines the process of determining precise target candidate boxes by modulating the confidence level, thereby enhancing detection accuracy. Finally, the focal loss function is introduced to train the model, reducing the weight of simple samples during the training process, and allowing the model to focus more on samples that are difficult to distinguish. Experimental results demonstrate that the enhanced C-SwinFaster algorithm significantly improves the identification accuracy of male and female Chinese mitten crab juveniles. The mean average precision (mAP) of this algorithm reaches 98.45%, marking a 10.33 percentage point increase over the original model. This algorithm has a good effect on the gender recognition of Chinese mitten crab juveniles and can provide technical support for the automatic classification of Chinese mitten crab juveniles.

Detection of variety and wax bloom of Shaanxi plum during post-harvest handling

The detection of plum variety and wax bloom has extensive applications in the fields of fruit classification and fruit quality assessment. By automating the detection and identification of plum varieties and wax bloom, it is possible to enhance the efficiency and accuracy of variety identification and quality assessment, and reduce manual intervention and misjudgment, thereby improving the market competitiveness of fruits. Currently, many works focus on improving the detection performance of single attribute detection of plum varieties or wax bloom, and it is often necessary to use two models to detect the same plum variety and quality information separately, which leads to inefficient and resource-consuming problems in practical applications. To solve this problem and improve the efficiency of detection, a Multi-Label detection model based on YOLOv7 is proposed. Firstly, the double detection head structure is introduced to improve the prediction ability for two types of attribute features. Then, the loss function suitable for multi-attribute labels is improved, and two classification loss functions are used to optimize the prediction results of the two types of attribute labels, respectively. Finally, a multi-label non-maximum suppression algorithm is proposed to solve the problem of filtering redundant bounding boxes of multi-attribute labels. Experimental results on the plum image dataset show that the proposed Multi-Label YOLOv7 model achieves a mAP@0.5 of 96.2 %, a precision of 94.6 %, and a recall of 89.5 %. The experimental results show that the Multi-Label YOLOv7 model can effectively detect variety and wax bloom attributes and improve the efficiency of multi-attribute label detection. The code and dataset for this experiment can be found at https://github.com/hejinrong/Muti-Label-YOLOv7.

High Throughput FPGA-Based Object Detection via Algorithm-Hardware Co-Design

Object detection and classification is a key task in many computer vision applications such as smart surveillance and autonomous vehicles. Recent advances in deep learning have significantly improved the quality of results achieved by these systems, making them more accurate and reliable in complex environments. Modern object detection systems make use of lightweight convolutional neural networks (CNNs) for feature extraction, coupled with single-shot multi-box detectors (SSDs) that generate bounding boxes around the identified objects along with their classification confidence scores. Subsequently, a non-maximum suppression (NMS) module removes any redundant detection boxes from the final output. Typical NMS algorithms must wait for all box predictions to be generated by the SSD-based feature extractor before processing them. This sequential dependency between box predictions and NMS results in a significant latency overhead and degrades the overall system throughput, even if a high-performance CNN accelerator is used for the SSD feature extraction component. In this paper, we present a novel pipelined NMS algorithm that eliminates this sequential dependency and associated NMS latency overhead. We then use our novel NMS algorithm to implement an end-to-end fully pipelined FPGA system for low-latency SSD-MobileNet-V1 object detection. Our system, implemented on an Intel Stratix 10 FPGA, runs at 400 MHz and achieves a throughput of 2,167 frames per second with an end-to-end batch-1 latency of 2.13 ms. Our system achieves 5.3× higher throughput and 5× lower latency compared to the best prior FPGA-based solution with comparable accuracy.

Refined fire detection and band selection method in hyperspectral remote sensing imagery based on sparse-VIT

Hyperspectral remote sensing has powerful ability to perceive information of land surface, and can be applied to the field of fire monitoring to achieve refined fire analysis. In this paper, we study the fire detection method based on hyperspectral remote sensing, design an end to end fire detection model based on sparse visual transformer, and propose a sensitive band selection method under the transformer framework. In order to suppress the interference of invalid bands in hyperspectral data, the sparse attention mechanism and the top-k selection mechanism are introduced in the model to zeroing the attention score of invalid bands. To achieve dimension reduction, a non-maximum attention suppression algorithm under the transformer framework is designed and band pruning is further introduced to eliminate invalid bands and redundant bands. The model adopts input mode of band-exclusive-token, which makes the pruning operation equivalent to band selection. We publish a hyperspectral fire detection dataset and the performance of our proposed model is verified on the dataset.

An Improved Lightweight YOLO Algorithm for Recognition of GPS Interference Signals in Civil Aviation

Considering several sources that cause global position system (GPS) interference in civil aviation and the challenges faced by interference recognition algorithms in terms of efficiency and accuracy, we propose an improved You Only Look Once (YOLO)v7‐CHS algorithm (YOLOv7‐CHS) and investigate its effectiveness in identifying GPS signals and different types of interference signals. First, continuous wavelet transform (CWT) is introduced as a method for processing and analyzing signals in the time–frequency (TF) domain to effectively obtain their temporal and spectral characteristic information. Second, the ConvNeXt structure is integrated into the YOLOv7 backbone network to create a ConvNeXtBlock (CNeB) module to enhance the classification and recognition accuracy of interference signals. Additionally, an attention mechanism is introduced to further improve model recognition accuracy. To effectively improve the capability of signal feature extraction and mitigate the impact of background noise on TF feature suppression, we have integrated the efficient channel attention (ECA) channel attention module with the convolutional block attention module (CBAM) spatial attention module, thereby proposing a hybrid CBAM and ECA (HCE) attention module. Last, to address issues arising from accidental deletion of detection frames and multipath interference negatively affecting model recognition performance, we have employed the soft nonmaximum suppression (Soft‐NMS) algorithm while selecting an optimal loss function through comparative analysis. The comparative evaluation experimental results under different circumstances show that YOLOv7‐CHS achieves recognition accuracies of 98.0% and 99.6% for various types of signals, respectively. These values represent an increase of 1.7% and 1%, respectively, compared to YOLOv7. Moreover, in terms of lightweight indicators, YOLOv7‐CHS exhibits a significant improvement in performance: the frames per second (FPS) is increased by 75.1, the number of parameters (Params) was reduced by 4.75 M, and giga floating point operations per second (GFLOPs) were reduced by 65.9 G while effectively enhancing recognition capabilities. The proposed YOLOv7‐CHS not only improves signal recognition accuracy but also reduces model Params and computational complexity, achieving a lightweight model with promising application prospects in the rapid detection and recognition of GPS interference sources in civil aviation.

SDMSEAF-YOLOv8: a framework to significantly improve the detection performance of unmanned aerial vehicle images

The detailed, high-resolution images captured by drones pose challenges to target detection algorithms with complex scenes and small-sized targets. Moreover, targets in unmanned aerial vehicle images are usually affected by factors such as viewing perspective, occlusion, and light, which increase the difficulty of target detection. In response to the above issues, we propose an improved SDMSEAF-YOLOv8 for target detection based on YOLOv8, combined with a Bi-directional Feature Pyramid Network, to improve the sensing ability of the model for multiscale targets. A Space-to-depth layer replaces the traditional strided convolution layer to enhance the extraction of fine-grained information and small-sized target features. A Multi-Separated and Enhancement Attention module enhances the feature learning ability of the occluded target region, thus reducing missed and false detections. Four detection heads are employed for tiny target detection, each responsible for different size ranges, so as to improve the accuracy and robustness of small target detection. The conventional non-maximum suppression algorithm is improved, so as to reduce the problem of missed detections under a densely occluded scene by setting the attenuation function to adjust the confidence of the treated box based on the overlap between it and the highest-scoring box. Experiments demonstrate that the accuracy of SDMSEAF-YOLOv8 exceeds that of state-of-the-art models on the VisDrone2019-DET-val dataset, with a mAP of 42.9% at 640-pixel resolution, 14.8% over the baseline YOLOv8-x algorithm model, and 6.0% over the known state-of-the-art Fine-Grained Target Focusing Network model and with twice as fast detection.

Attention Mechanism and Detection Box Information Based Real-time Multi-Object Vehicle Detection

Ensuring both the accuracy of vehicle target detection and meeting real-time requirements is crucial in traffic videos. The YOLOv5s target detection framework, known for its accuracy and efficiency, has attracted attention in academic circles. However, there are still some features that can be optimized. First of all, the detection subnet of the YOLOv5s framework cannot smoothly convert complex feature maps into relatively sparse target prediction boxes. To solve this, we integrate a self-attention-based gating mechanism into the detection subnet, forming the YOLOv5s-SAG network. Secondly, the loss function of CIoU used by YOLOv5s pays insufficient attention to the overlapping area of the detection frame, which can be used as metric for measuring target detection effectiveness. We add the loss term of area ratio to CIoU to further improve the modeling ability. Finally, the current multi-class Non-Maximum Suppression algorithm can cause high overlap of multi-class detection frames. To improve it, we propose a multi-class CS-NMS algorithm based on category suppression. Experimental results show an approximately 8% improvement in the mAP50 index on the UA-DETRAC dataset compared with YOLOv5s. The proposed algorithm also achieves better detection results compared to mainstream target detection algorithms and meets the real-time requirements of traffic video analysis.