Deep Learning Object Detection Research Articles

Telecommunication Fiber Box Detection Using YOLO in Urban Environment

The Fiber Distribution Panel (FDP) box is an essential piece of internet access hardware because it provides users with high-speed data networking and functions as a cable organizer to reduce wire clutter. After installing the FDP, an inspection must be performed to ensure that all necessary components are present. However, This examination is still done manually; the technician snaps a picture of the panel and sends it to its supervisor for verification, which is time-consuming and often prone to errors. In addition to images captured in low-light and complex environments, it makes it more difficult for humans to identify the components with just a naked eye. On this matter, a much more efficient method to assess the FDP installation work is very much needed. Therefore, using computer vision approaches, we utilize a deep learning algorithm to perform object detection and automate the assessment of FDP installation components based on visual data. One of the deep learning models established in the literature is the You Only Look Once (YOLO) model, a one-stage deep learning object detection algorithm that employs a fully conventional approach to generate highly accurate real-time predictions. This paper uses YOLOv5s to identify the fiber box and its relevant components, even in urban environments. Experimentations show that YOLO successfully identified the installation parts with a mean average precision score of 86% at a 0.5 confidence level, even with limited data.

Vegetation and Non-Vegetation Classification Using Object Detection Techniques and Deep Learning from Low/Mixed Resolution Satellite Images

Vegetation cover classification using mixed or low-resolution scalar images is challenging. Fortunately, recently deep learning object detection methods have emerged as a replacement to the conventional machine learning methods for the detection and classification of land use and land cover. This paper presents a deep learning object detection approach for land use and land cover detection using low/mixed resolution satellite images acquired from Google Earth satellite images. Google Earth images are accessible freely using the Google Earth Pro desktop application. Our dataset consists of two (02) classes (vegetation and non-vegetation) with a total of 450 labeled images captured from different parts of Pakistan. We present a comparison of the recent anchor-free object detection model YOLOX with the anchor-based object detection model YOLOR for solving real-time problems. The end-to-end differentiability, efficient GPU utilization, and absence of hand-crafted parameters make anchor-free models a compelling choice in object detection, and yet not been explored on Land cover classification using satellite images. Our experimental study shows that YOLOX delivers an overall accuracy of 83.50% on Vegetation and 86% on Non-Vegetation classes, which outperformed YOLOR by 30% on Vegetation classes and 34% on non-Vegetation classes for our dataset. We also show how an object detection system can be used for Vegetation and Non-Vegetation classification tasks, which can then be used for change monitoring and assisting in developing geographical maps using low/mixed resolution freely available satellite images.

A Comparative Analysis of Single-Stage Detectors from the Perspectives of Anchor-Free and Anchor-Based Approaches

Since the emergence of deep learning in object detection in 2014, there has been rapid development and widespread application of detectors across various domains such as healthcare, road traffic, and industrial inspection. The evolution of convolutional neural network (CNN)-based detectors has reached a mature stage, with single-stage detectors playing a crucial role within CNN-based frameworks. This paper provides an analysis of the developmental process and distinctive features of classic single-stage detectors categorized into two types: Anchor-free and Anchor-based. Furthermore, the paper also provides a comparison of the accuracy trends for these two types of single-stage detectors over different years. Through this analysis and comparison, potential reasons behind the advancements in single-stage detectors during different time periods can be explored, such as network architecture optimization, increase in datasets, and algorithm improvements. Additionally, this paper offers insights into current research focal points and future prospects.

PROCEEDINGS IN UAS-ASSISTED BRIDGE INSPECTIONS: RTK-BASED PHOTOGRAMMETRIC RECONSTRUCTION AND SPATIAL FILTERING

Abstract. Traditional bridge inspections present considerable challenges in terms of efficiency and accuracy. However, recent advancements in Unmanned Aerial Systems (UASs) and deep learning for object detection have opened up new avenues for automatic bridge damage detection. We present a comprehensive framework leveraging these technologies for automated damage detection in UAS imagery, followed by accurate mapping of the damage predictions on photogrammetric models. In this work, we propose a photogrammetric procedure to retrieve geolocated bridge models solely based on Real-Time Kinematics (RTK) information. Within the damage detection step, we conduct extensive testing and optimization of model hyperparameters using YOLOv8 and Slicing Aided Hyper Interference (SAHI). Next, we map the predictions onto the 3D model using ray casting, allowing to group and filter the predictions by their area and position. Finally, a Graphical User Interface (GUI) allows bridge inspectors to identify false positive predictions, generate new training data, and directly measure damage dimensions in the images. Validation on a concrete box girder bridge resulted in a photogrammetric model with a mean error of 1.3 cm, negating the need for ground control points. Our model training process revealed substantial performance variations between training and test datasets, underscoring the importance of evaluating optimal hyperparameters on UAS inspection images rather than relying on the validation metrics. Lastly, we successfully map the detected damage and create new training data from the UAS inspection images. This framework significantly enhances bridge inspection accuracy and efficiency, providing a strong foundation for future developments in automated bridge inspections.

Visual Sensing and Depth Perception for Welding Robots and Their Industrial Applications.

With the rapid development of vision sensing, artificial intelligence, and robotics technology, one of the challenges we face is installing more advanced vision sensors on welding robots to achieve intelligent welding manufacturing and obtain high-quality welding components. Depth perception is one of the bottlenecks in the development of welding sensors. This review provides an assessment of active and passive sensing methods for depth perception and classifies and elaborates on the depth perception mechanisms based on monocular vision, binocular vision, and multi-view vision. It explores the principles and means of using deep learning for depth perception in robotic welding processes. Further, the application of welding robot visual perception in different industrial scenarios is summarized. Finally, the problems and countermeasures of welding robot visual perception technology are analyzed, and developments for the future are proposed. This review has analyzed a total of 2662 articles and cited 152 as references. The potential future research topics are suggested to include deep learning for object detection and recognition, transfer deep learning for welding robot adaptation, developing multi-modal sensor fusion, integrating models and hardware, and performing a comprehensive requirement analysis and system evaluation in collaboration with welding experts to design a multi-modal sensor fusion architecture.

Deep Learning-Based Oyster Packaging System

With the deepening understanding of the nutritional value of oysters by consumers, oysters as high-quality seafood are gradually entering the market. Raw edible oyster production lines mainly rely on manual sorting and packaging, which hinders the improvement of oyster packaging efficiency and quality, and it is easy to cause secondary oyster pollution and cross-contamination, which results in the waste of oysters. To enhance the production efficiency, technical level, and hygiene safety of the raw aquatic products production line, this study proposes and constructs a deep learning-based oyster packaging system. The system achieves intelligence and automation of the oyster packaging production line by integrating the deep learning algorithm, machine vision technology, and mechanical arm control technology. The oyster visual perception model is established by deep learning object detection techniques to realize fast and real-time detection of oysters. Using a simple online real-time tracking (SORT) algorithm, the grasping position of the oyster can be predicted, which enables dynamic grasping. Utilizing mechanical arm control technology, an automatic oyster packaging production line was designed and constructed to realize the automated grasping and packaging of raw edible oysters, which improves the efficiency and quality of oyster packaging. System tests showed that the absolute error in oyster pose estimation was less than 7 mm, which allowed the mechanical claw to consistently grasp and transport oysters. The static grasping and packing of a single oyster took about 7.8 s, and the success rate of grasping was 94.44%. The success rate of grasping under different transportation speeds was above 68%.

Identification of the normative use of medical protective equipment by fusion of object detection and keypoints detection

Background and ObjectiveIrregular use of personal protective equipment (PPE) seriously affects the occupational health and safety of healthcare workers, especially in large public health emergencies such as COVID-19. The danger often occurs in some complex scenarios where the use of certain equipment does not comply with the spatial rules it specifies. The single object detection-based method is difficult to effectively verify whether the worker's use of PPE is normative. Also detecting the use of six classes of PPE brings a large computational load to the task. MethodsIn this paper, we proposed an identification approach that combined human keypoints detection with deep learning object detection to help facilitate the monitoring of healthcare workers' standard PPE use. We used YOLOv4 as the baseline model for PPE detection and MobileNetv3 as the backbone of the detector to reduce the computational effort. In addition, High-Resolution Net (HRNet) was the benchmark for keypoints detection, characterizing the coordinates of 25 key points in the human body. Generalized Intersection over Union (GIoU) was used to establish the association between PPEs and key points, and by calculating the matching score between a PPE and the corresponding body bounding boxes, the authenticity of the PPE specification could be effectively inferred. ResultsThe proposed detector is able to identify whether a healthcare worker is normatively using multiple equipment with a higher precision (95.81 %), recall (96.38 %), and F1-score (96.09 %). Meanwhile, the number of parameters (2.87 M) and the size of the model (6.4 MB) are also more lightweight than other comparative detectors. ConclusionsOur approach is more reliable for reasoning about the normality of personal protection for healthcare workers in some complex scenarios than a single object detection-based approach. The developed identification framework provides a new automated monitoring solution for protection management in healthcare, and the modular design brings more flexible applications for different medical operation scenarios.

Lightweight Deep Learning for Object Detection on Mobile Device

Computer vision is a research in the development of technology to obtain information from images and replicate or imitate human visual processes, so that computers can know the objects around them. Deep learning is now the key word as a new era in machine learning that trains computers in finding patterns from large amounts of data. The Convolution Neural Networks (CNN) algorithm has proven impressive in terms of performance for detecting objects, image classification and semantic segmentation. Object detection is a technique used to identify the type of object in an image and also the exact location of the object in the image. Face detection is one of the most challenging problems of pattern recognition. Effective training needs to be done to be able to detect faces effectively. The accuracy in face detection using machine learning does not give good results. This research focuses on the level of accuracy of detecting faces using deep learning methods. This study compares the level of accuracy of deep learning and machine learning in detecting faces effective and efficient. This study uses the Convolution Neural Networks (CNN) model in the deep learning method to detect faces in real time on Android. According to the test results, the accuracy obtained in this study reached 97.97% in several normal facial conditions and face masks.

넙치 질병 증상 분류를 위한 객체 탐지 딥러닝 모델 성능 평가

Paralichthys olivaceus accounts for a large proportion, accounting for more than half of Korea's aquaculture industry. However, about 25-30% of the total breeding volume throughout the year occurs due to diseases, which has a very bad impact on the economic feasibility of fish farms. For the economic growth of Paralichthys olivaceus farms, it is necessary to quickly and accurately diagnose disease symptoms by automating the diagnosis of Paralichthys olivaceus diseases. In this study, we create training data using innovative data collection methods, refining data algorithms, and techniques for partitioning dataset, and compare the Paralichthys olivaceus disease symptom detection performance of four object detection deep learning models(such as YOLOv8, Swin, Vitdet, MvitV2). The experimental findings indicate that the YOLOv8 model demonstrates superiority in terms of average detection rate (mAP) and Estimated Time of Arrival (ETA). If the performance of the AI model proposed in this study is verified, Paralichthys olivaceus farms can diagnose disease symptoms in real time, and it is expected that the productivity of the farm will be greatly improved by rapid preventive measures according to the diagnosis results

Comparison of Cloud-Computing Providers for Deployment of Object-Detection Deep Learning Models

As cloud computing rises in popularity across diverse industries, the necessity to compare and select the most appropriate cloud provider for specific use cases becomes imperative. This research conducts an in-depth comparative analysis of two prominent cloud platforms, Microsoft Azure and Amazon Web Services (AWS), with a specific focus on their suitability for deploying object-detection algorithms. The analysis covers both quantitative metrics—encompassing upload and download times, throughput, and inference time—and qualitative assessments like cost effectiveness, machine learning resource availability, deployment ease, and service-level agreement (SLA). Through the deployment of the YOLOv8 object-detection model, this study measures these metrics on both platforms, providing empirical evidence for platform evaluation. Furthermore, this research examines general platform availability and information accessibility to highlight differences in qualitative aspects. This paper concludes that Azure excels in download time (average 0.49 s/MB), inference time (average 0.60 s/MB), and throughput (1145.78 MB/s), and AWS excels in upload time (average 1.84 s/MB), cost effectiveness, ease of deployment, a wider ML service catalog, and superior SLA. However, the decision between either platform is based on the importance of their performance based on business-specific requirements. Hence, this paper ends by presenting a comprehensive comparison based on business-specific requirements, aiding stakeholders in making informed decisions when selecting a cloud platform for their machine learning projects.

Anomalous behavior recognition of underwater creatures using lite 3D full-convolution network

Global warming and pollution could lead to the destruction of marine habitats and loss of species. The anomalous behavior of underwater creatures can be used as a biometer for assessing the health status of our ocean. Advances in behavior recognition have been driven by the active application of deep learning methods, yet many of them render superior accuracy at the cost of high computational complexity and slow inference. This paper presents a real-time anomalous behavior recognition approach that incorporates a lightweight deep learning model (Lite3D), object detection, and multitarget tracking. Lite3D is characterized in threefold: (1) image frames contain only regions of interest (ROI) generated by an object detector; (2) no fully connected layers are needed, the prediction head itself is a flatten layer of 1 × {mathcal{l}} @ 1× 1, {mathcal{l}}= number of categories; (3) all the convolution kernels are 3D, except the first layer degenerated to 2D. Through the tracking, a sequence of ROI-only frames is subjected to 3D convolutions for stacked feature extraction. Compared to other 3D models, Lite3D is 50 times smaller in size and 57 times lighter in terms of trainable parameters and can achieve 99% of F1-score. Lite3D is ideal for mounting on ROV or AUV to perform real-time edge computing.

Automatic detection of charcoal kilns on Very High Resolution images with a computer vision approach in Somalia

In Somalia’s drylands, charcoal production is a major driver of forest degradation enabled by civil conflict and institutional weakness. Up to now, the extent and exact location of charcoal production has usually been estimated by visually detecting charcoal kilns on Very High Resolution (VHR) images. Taking advantage of the availability of a dataset of charcoal kilns delineated on VHR images by the Food and Agriculture Organization (FAO) experts, we designed a computer vision (CV) approach to automatically identify charcoal kilns on VHR images. The methodology relies on a curated subset of the expert-labeled dataset and a collection of panchromatic and pan-sharpened multispectral Natural Color (RGB) VHR for the years 2018 and 2019. Kiln delineations paired with the VHR images are visually reviewed and used with a Faster R-CNN model, an object detection deep learning method. A two-stage methodology is used to train the best models for panchromatic and RGB images, respectively. The first stage uses a small number of high quality pairs to define the best parameters of the models while the second stage uses a larger set of pairs to fine tune the previous models. The results indicates that charcoal kilns are detected with a precision of 90% in panchromatic images and of 80% in RGB images. The models are then used to predict the presence of kilns over the available VHR images. Omission errors are prioritized over commission errors to mitigate the difficulty in detecting kilns in some specific situations. Comparison between the FAO expert kiln dataset and objects predicted by the CNN model is giving encouraging results. With a visual screening complementing the proposed workflow, CV can aid charcoal kiln monitoring in Somalia while alleviating manual work.

Automated Image Quality and Protocol Adherence Assessment of Examinations in Teledermatology: First Results.

Introduction: Image quality and acquisition protocol adherence assessment is a neglected area in teledermatology. We examine if it is feasible to use deep learning methods to automate the assessment of the adherence of examinations to image acquisition protocols. In this study, we focused on the quality criteria of two image acquisition protocols: (1) approximation image and (2) panoramic image, as these are present in all teledermatology examination protocols currently used by the Santa Catarina State Integrated Telemedicine and Telehealth System (STT/SC). Methods: We use a data set of 36,102 teledermatological examinations performed at the STT/SC during 2021. As our validation process, we adopted standard machine learning metrics and an inter-rater agreement (IRA) study with 11 dermatologists. For the approximation image protocol, we used the Mask-Region based Convolutional Neural Network (RCNN) Object Detection Deep Learning (DL) architecture to identify the presence of a lesion identification tag and a ruler used to provide a frame reference of the lesion. For the panoramic image protocol, we used DensePose, a pose estimation DL, architecture to assess the presence of a whole patient body and its orientation. A combination of the two approaches was additionally validated through an IRA study between specialists. Results: Mask-RCNN achieved a score of 96% mean average precision (mAP), while DensePose presented 75% mAP. IRA achieved a level of agreement of 96.68% with the Krippendorff alpha score. Conclusions: Our results show the feasibility of using deep learning to automate the image quality and protocol adherence assessment in teledermatology, before the specialist's manual analysis of the examination.

Automated Leopard Alert And Reporting Mechanism Using Deep Learning

Today, rapid infrastructure development is taking place in major metropolitan cities, but unfortunately, this progress often involves the destruction of forest reserves, leaving wild animals homeless. The resulting environmental invasion forces these animals to venture into the cities, posing threats to citizens. In Mumbai, there have been numerous sightings of leopards and other wild animals near forested areas. Leopards have been known to attack street dogs, people, and vehicles, making it necessary to work on this problem. This paper suggests the utilization of deep learning models and object detection techniques to detect leopards and other potential threats. By integrating this technology with security applications, citizens can be made aware of the existence of wild animals in their vicinity. This research primarily focuses on addressing the concern of leopard sightings in Mumbai. The objective is to automate leopard detection and reporting using an object detection algorithm. In the proposed system, images of leopards are collected from an existing dataset available on Roboflow, comprising a total of 1000 samples. The proposed model's performance is evaluated using Mean Average Precision (mAP) & detection speed. The proposed method achieves an impressive mAP of 95.9% at a speed of 37 frames per second.

Review of Object Detection Algorithms for Sonar Images based on Deep Learning

Background: Deep learning object detection algorithm is widely used in the field of image classification and has become an indispensable part. With the improvement of image classification accuracy, sonar image target detection algorithm based on deep learning has gradually become the focus of more and more people's research. Objective: This article aims to provide a summary and analysis of deep learning-based sonar image object detection algorithms, with the hope of offering insights for future research in the field of sonar target detection technology. Method: This paper systematically summarizes sonar image target detection algorithms based on deep learning. According to the method principle, the existing deep learning target detection algorithms are divided into four categories: target detection algorithm based on candidate region, deep target detection method based on regression, Anchor Free deep learning target detection algorithm, and search-based target detection and recognition algorithm. Then, the performance of algorithms based on COCO data sets is compared, and the standard sonar data sets and formats are introduced. Results: The sonar image object detection algorithm based on deep learning has made significant progress. The combination of deep learning and object detection methods has been applied to sonar images, resulting in the emergence of excellent performing algorithms. However, most algorithms are still in the developmental stage and face challenges in practical applications. Subsequently, several invention patents have been developed based on the aforementioned algorithms, including a feature extraction method for side-scan sonar images based on fully convolutional neural networks, an underwater sonar image target detection method based on improved YOLOv3-tiny, and more. Conclusion: Sonar image object detection technology based on deep learning has a wide range of application needs but also faces many difficulties and challenges, we still need to continue to learn and explore in future research, and we believe that we can make greater breakthroughs in the future.

An approach to ship target detection based on combined optimization model of dehazing and detection

The design of a ship detection model that can be adapted to both foggy and clear images faces significant challenges. Existing methods are either not accurate enough, or have a high amount of model parameters, making them difficult to deploy to lightweight front-ends. To address these issues, a lightweight deep learning model based on combined optimization of dehazing and detection is proposed, focusing on self-adaptive ship detection. Firstly, a self-adaptive image dehazing module is designed and placed ahead of the detection network, including a dehazing parameter predictor and an improved dehazing method. Subsequently, a lightweight-improved object detection deep learning model integrated with the dehazing module is devised to detect the ship in the foggy image. Experimental results demonstrate the effectiveness of this approach in enabling efficient and accurate ship detection under foggy conditions. Through the joint optimization of the dehazing module and the detection module, it can be seen from the experiments that our Dehazing + Detection model has the highest detection accuracy and performs well in terms of detection speed, parameter amount, and weight file size. The detection accuracy has reached 97.1%, which is better than that of the other three dehazing + detection models.

A Task-Risk Consistency Object Detection Framework Based on Deep Reinforcement Learning

A discernible gap has materialized between the expectations for object detection tasks in optical remote sensing images and the increasingly sophisticated design methods. The flexibility of deep learning object detection algorithms allows the selection and combination of multiple basic structures and model sizes, but this selection process relies heavily on human experience and lacks reliability when faced with special scenarios or extreme data distribution. To address these inherent challenges, this study proposes an approach that leverages deep reinforcement learning within the framework of vision tasks. This study introduces a Task-Risk Consistent Intelligent Detection Framework (TRC-ODF) for object detection in optical remote sensing images. The proposed framework designs a model optimization strategy based on deep reinforcement learning that systematically integrates the available information from images and vision processes. The core of the reinforcement learning agent is the proposed task-risk consistency reward mechanism, which is the driving force behind the optimal prediction allocation in the decision-making process. To verify the effectiveness of the proposed framework, multiple sets of empirical evaluations are conducted on representative optical remote sensing image datasets: RSOD, NWPU VHR-10, and DIOR. When applying the proposed framework to representative advanced detection models, the mean average precision (mAP@0.5 and mAP@0.5:0.95) is improved by 0.8–5.4 and 0.4–2.7, respectively. The obtained results showcase the considerable promise and potential of the TRC-ODF framework to address the challenges associated with object detection in optical remote sensing images.

Obstacle Avoidance for Automated Guided Vehicles in Real-World Workshops Using the Grid Method and Deep Learning

An automated guided vehicle (AGV) obstacle avoidance system based on the grid method and deep learning algorithm is proposed, aiming at the complex and dynamic environment in the industrial workshop of a tobacco company. The deep learning object detection is used to detect obstacles in real-time for the AGV, and feasible paths are generated by the grid method, which ultimately finds an AGV obstacle avoidance solution in complex dynamic environments. The experimental results showed that the proposed system can effectively identify and avoid obstacles in a simulated tobacco production workshop environment, resulting in the average obstacle avoidance success rate of 98.67%. The transportation efficiency of cigarette factories is significantly improved with the proposed system, reducing the average execution time of handing tasks by 27.29%. This paper expects to provide a reliable and efficient solution for AGV obstacle avoidance in real-world industrial workshops.

Deep Neural Network-Based Automatic Dicentric Chromosome Detection Using a Model Pretrained on Common Objects.

Dicentric chromosome assay (DCA) is one of the cytogenetic dosimetry methods where the absorbed dose is estimated by counting the number of dicentric chromosomes, which is a major radiation-induced change in DNA. However, DCA is a time-consuming task and requires technical expertise. In this study, a neural network was applied for automating the DCA. We used YOLOv5, a one-stage detection algorithm, to mitigate these limitations by automating the estimation of the number of dicentric chromosomes in chromosome metaphase images. YOLOv5 was pretrained on common object datasets. For training, 887 augmented chromosome images were used. We evaluated the model using validation and test datasets with 380 and 300 images, respectively. With pretrained parameters, the trained model detected chromosomes in the images with a maximum F1 score of 0.94 and a mean average precision (mAP) of 0.961. Conversely, when the model was randomly initialized, the training performance decreased, with a maximum F1 score and mAP of 0.82 and 0.873%, respectively. These results confirm that the model could effectively detect dicentric chromosomes in an image. Consequently, automatic DCA is expected to be conducted based on deep learning for object detection, requiring a relatively small amount of chromosome data for training using the pretrained network.

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.