YOLOv3 Model Research Articles

Deep Neural Network Driven Automated Underwater Object Detection

Object recognition and computer vision techniques for automated object identification are attracting marine biologist's interest as a quicker and easier tool for estimating the fish abundance in marine environments. However, the biggest problem posed by unrestricted aquatic imaging is low luminance, turbidity, background ambiguity, and context camouflage, which make traditional approaches rely on their efficiency due to inaccurate detection or elevated false-positive rates. To address these challenges, we suggest a systemic approach to merge visual features and Gaussian mixture models with You Only Look Once (YOLOv3) deep network, a coherent strategy for recognizing fish in challenging underwater images. As an image restoration phase, pre-processing based on diffraction correction is primarily applied to frames. The YOLOv3 based object recognition system is used to identify fish occurrences. The objects in the background that are camouflaged are often overlooked by the YOLOv3 model. A proposed Bi-dimensional Empirical Mode Decomposition (BEMD) algorithm, adapted by Gaussian mixture models, and integrating the results of YOLOv3 improves detection efficiency of the proposed automated underwater object detection method. The proposed approach was tested on four challenging video datasets, the Life Cross Language Evaluation Forum (CLEF) benchmark from the F4K data repository, the University of Western Australia (UWA) dataset, the bubble vision dataset and the DeepFish dataset. The accuracy for fish identification is 98.5 percent, 96.77 percent, 97.99 percent and 95.3 percent respectively for the various datasets which demonstrate the feasibility of our proposed automated underwater object detection method.

Deep Learning Based Audio Assistive System for Visually Impaired People

Vision impairment is a latent problem that affects numerous people across the globe. Technological advancements, particularly the rise of computer processing abilities like Deep Learning (DL) models and emergence of wearables pave a way for assisting visually-impaired persons. The models developed earlier specifically for visually-impaired people work effectually on single object detection in unconstrained environment. But, in real-time scenarios, these systems are inconsistent in providing effective guidance for visually-impaired people. In addition to object detection, extra information about the location of objects in the scene is essential for visually-impaired people. Keeping this in mind, the current research work presents an Efficient Object Detection Model with Audio Assistive System (EODM-AAS) using DL-based YOLO v3 model for visually-impaired people. The aim of the research article is to construct a model that can provide a detailed description of the objects around visually-impaired people. The presented model involves a DL-based YOLO v3 model for multi-label object detection. Besides, the presented model determines the position of object in the scene and finally generates an audio signal to notify the visually-impaired people. In order to validate the detection performance of the presented method, a detailed simulation analysis was conducted on four datasets. The simulation results established that the presented model produces effectual outcome over existing methods.

Remote Sensing Image Object Detection Method Based On Improved YOLOv3

Abstract In order to solve the problem that irregular targets and dense targets are difficult to be detected in optical remote sensing images, this paper improved the YOLOV3 model Firstly, in order to further combine the feature information of different scales, the PaNet structure is introduced into the FPN part of the original YOLOv3, and the obtained effective feature layer is continued to be extracted for a round of feature. The feature is not only up-sampled to achieve feature fusion, but also down-sampled again to achieve enhanced feature fusion SimOTA method is introduced to dynamically match positive samples and set different positive sample numbers for different targets, which not only improves the speed of the algorithm, but also reduces the extra hyperparameters Experimental verification using richer DOIR data sets shows that the detection ability of the improved algorithm is significantly improved. Compared with the original YOLOv3, its mAP improves by 15.1 points, among which the detection accuracy of dense small targets is improved the most.

Automatic body region localization in 3D-CT images based on the improved YOLO model

Automatic body region localization in medical three-dimensional (3D)-CT images is a critical step of computerized body-wide Automatic Anatomy Recognition (AAR) system, which can be applied for radiotherapy planning and interest slices retrieving. Currently, the complex internal structure of human body and time consuming computation are the main challenges for the localization. Therefore, this paper introduces and improves the YOLO-v3 model into the body region localization for these problems. First, seven categories of body regions in a CT volume image I are defined based on the modification version of our previous work. Second, an improved YOLO-v3 model is trained to classify each axial slice into one of the seven categories. Then, the effectiveness of the proposed method is evaluated on 3D-CT images that collected from 220 subjects. The experimental results demonstrate that the slice localizing error is less than 3 NoS (Number of slices), which is competitive to the state-of-the-art methods. Beyond this, our method is simple and computationally efficient owing to its less training time, and the average computational time for localizing a volume CT images is about 3 second, which shows potential for a further application.

Autonomous Recognition of Multiple Surgical Instruments Tips Based on Arrow OBB-YOLO Network

In surgery, the robot can automatically adjust the endoscope to ensure the stability of the surgical image and reduce the burden on the doctor. It is becoming a research hotspot of current surgical assistance. The identification and localization of endoscopic instruments tips is the key to achieve efficient “doctor–robot” collaboration. However, traditional methods have the problems of poor real-time performance and insufficient information expression. This article proposes an autonomous recognition of multiple surgical instrument tips based on arrow object bounding box (OBB)-YOLO network prediction, the proposed method not only realizes the recognition and localization of key parts of the instrument, but also uses arrows to enhance the expression of critical information such as the tip angle under the premise of meeting the real-time requirements for medical scenes. First, based on the modified YOLOv3 model, the Arrow OBB-YOLO network used to predict Arrow OBB is built, and the arrows are constrained inside the bounding box (BB) through coordinate transformation. Second, the optimization expression of arrow errors can enhance the prediction effect of the model, and the designed step-by-step training optimization method significantly accelerates the convergence speed of the model. Then, based on the Arrow OBB model, a method for generating the tracking areas is proposed. Finally, the experimental results show that the proposed method can accurately identify and locate the tips of the instrument with 36.5 fps and can improve the recognition effect of the BB when adding new information expression, which lays the foundation for “doctor–robot” collaborative surgery.

A Two-Tier Framework Based on GoogLeNet and YOLOv3 Models for Tumor Detection in MRI

Medical Image Analysis (MIA) is one of the active research areas in computer vision, where brain tumor detection is the most investigated domain among researchers due to its deadly nature. Brain tumor detection in magnetic resonance imaging (MRI) assists radiologists for better analysis about the exact size and location of the tumor. However, the existing systems may not efficiently classify the human brain tumors with significantly higher accuracies. In addition, smart and easily implementable approaches are unavailable in 2D and 3D medical images, which is the main problem in detecting the tumor. In this paper, we investigate various deep learning models for the detection and localization of the tumor in MRI. A novel two-tier framework is proposed where the first tire classifies normal and tumor MRI followed by tumor regions localization in the second tire. Furthermore, in this paper, we introduce a well-annotated dataset comprised of tumor and normal images. The experimental results demonstrate the effectiveness of the proposed framework by achieving 97% accuracy using GoogLeNet on the proposed dataset for classification and 83% for localization tasks after fine-tuning the pre-trained you only look once (YOLO) v3 model.

Recognize highly similar sewing gestures by the robot

The autonomous and efficient learning of sewing gestures by robots will bring great convenience to the garment industry. To improve the accuracy of robots in detecting sewing gestures with high similarity, three detection models based on deep learning are proposed in the paper. First, in order to improve the detection accuracy and detection speed of sewing gestures under complex backgrounds, we added a dense connection layer to the low-resolution network layer of YOLO-V3 to enhance the transmission and reuse rate of image features. Secondly, a deeper ResNet50 residual network is introduced to replace the VGG16 basic network in the original SSD model. The feature pyramid structure is used to fuse high-level semantic features and low-level semantic features, which can improve the detection accuracy of small-sized sewing gestures. Finally, the parallel spatial-temporal dual-stream network separately extracts the temporal feature and the spatial feature of sewing gestures. The fusion of time feature and space feature improves the detection accuracy of the coherent sewing gesture. The results show that the suggested three models can effectively detect four sewing gestures with high similarity. Among them, the spatial-temporal two-stream convolutional neural network has the highest detection accuracy. The improved SSD model has faster detection speed than the improved YOLO-V3 model and other mainstream algorithms.

Identification and Depth Localization of Clustered Pod Pepper Based on Improved Faster R-CNN

Traditionally height of end effector of pod pepper harvester is fixed, which induces it hardly adapt to growth height of clustered peppers. Firstly, aiming at the problems of small size and clustered growth of pepper fruits during identification task, an improved Faster R-CNN algorithm is proposed. On the one hand, strategies such as increasing the types and number of high-resolution anchors and using RoI Align instead of RoI Pooling are employed to improve the detection accuracy for tiny targets. On the other hand, ResNet+FPN instead of VGG16 and ResNet backbone structure is adopted as the low-level feature extractor, so extracting capability for small features can be enhanced effectively. Furthermore, to precisely locate the position of clustered peppers, a height calculation model combining the 2D image recognition results with its depth information is advanced. Comparative experiments show that the overall accuracy <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">AP</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">AP</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</sub> of our method reach 75.79% and 87.30%, respectively. Compared with VGG16 feature extraction model, the two indicators are improved by 8.7% and 1.3%, respectively. The small target detection accuracy <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">AP</i> <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">small</sup> is increased about 11.4%, with recall rate <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">AR</i> <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">small</sup> increased up to 10.2%. The overall loss rate <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Loss</i> is reduced by 4.7%, which manifests greatly improvement compared to YOLOv3 model. The detection time of a single frame reaches 42 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ms</i> , which is slightly longer than that of YOLOv3 network, but it can still meet the real-time detection requirements of pepper harvester. In 3D location experiment, the average absolute height error of clustered peppers from the ground is 4.4 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">mm</i> , that accounts to the relative average error of 1.1%, thus suffices the adjustment error requirement of the end effector.

Litchi detection in the field using an improved YOLOv3 model

Due to the illumination, complex background, and occlusion of the litchi fruits, the accurate detection of litchi in the field is extremely challenging. In order to solve the problem of the low recognition rate of litchi-picking robots in field conditions, this study was inspired by the ideas of ResNet and dense convolution and proposed an improved feature-extraction network model named “YOLOv3_Litchi”, combining dense connections and residuals for the detection of litchis. Firstly, based on the traditional YOLOv3 deep convolution neural network and regression detection, the idea of residuals was to be put into the feature-extraction network to effectively avoid the problem of decreasing detection accuracy due to the excessive depths of the network layers. Secondly, under the premise of a good receptive field and high detection accuracy, the large convolution kernel was replaced by a small convolution kernel in the shallow layer of the network, thereby effectively reducing the model parameters. Finally, the idea of feature pyramid was used to design the network to identify the small target litchi to ensure that the shallow features were not lost and simultaneously reduced the model parameters. Experimental results show that the improved YOLOv3_Litchi model achieved better results than the classic YOLOv3_DarkNet-53 model and the YOLOv3_Tiny model. The mean average precision (mAP) score was 97.07%, which was higher than the 95.18% mAP of the YOLOv3_DarkNet-53 model and the 94.48% mAP of the YOLOv3_Tiny model. The frame frequency was 58 fps, which was higher than 29 fps of the YOLOv3_DarkNet-53 model. Compared with the classic Faster R-CNN model with the feature-extraction network VGG16, the mAP was increased by 1%, and the FPS advantage was obvious. Compared with the classic single shot multibox detector (SSD) model, both the accuracy and the running efficiency were improved. The results show that the improved YOLOv3_Litchi model had stronger robustness, higher detection accuracy, and less computational complexity for the identification of litchi in the field conditions, which should be helpful for litchi orchard precision management. Keywords: deep learning; residual network; dense connection; feature pyramid network DOI: 10.25165/j.ijabe.20221502.6541 Citation: Peng H X, Xue C, Shao Y Y, Chen K Y, Liu H N, Xiong J T, et al. Litchi detection in the field using an improved YOLOv3 model. Int J Agric & Biol Eng, 2022; 15(2): 211–220.

An Improved YOLOv3 for Foreign Objects Detection of Transmission Lines

The surrounding of transmission lines is complex, large-scale construction machinery, and smoke and fire pose threats to the circuit facilities and wires. When anomaly occurs, much time is required for the State Grid Corporation to fix it manually. To reduce the inspection burden, we propose a lightweight model running on embedded device to detect foreign objects of transmission lines. Based on the You Only Look Once (YOLO) v3, we use Mobilenetv2 instead of Darknet-53 as the backbone, and use depthwise separable convolution to replace 3 x 3 convolutional kernels in detection head, which greatly reduces the parameter size of the network. And the Fully Convolutional One-Stage Object Detection (FCOS)-like encoding and decode scheme is adopted to reduce network complexity. Meanwhile, in order to compensate for the degradation of accuracy, we have improved data augmentation, learning rate, and loss function. The experiments show that compared with other existed models, the improved YOLOv3 model has a smaller model size and higher detection speed without notably reducing detection accuracy, which has achieved the balance between detection speed and accuracy.

Waste Collection and Transportation Supervision Based on Improved YOLOv3 Model

Supervising the process of garbage collection and transportation is a very crucial task with significance implication to the reusing and the recycle of the garbage. Previous works mainly focus on the detection of certain kind of garbage instead of overseeing the whole process. This paper proposed a supervision approach based on the improved YOLOv3 to decrease the incidence of unexpected mixing of the different kinds of garbage during the transportation and the collection which will cause the inferior performance of waste classification. Firstly, to reduce the parameters and arithmetic operations of YOLOv3 and improving the model&#x2019;s detection speed, we displaced the standard convolution in YOLOv3 by depth-wise separable convolution. Secondly, to solve the problem that YOLOv3 has poor location accuracy and performs poorly in muti-targets, triplet attention is introduced into the backbone, which increases almost no parameters to automatically learn cross-dimensional interactions, enhance the effective feature channel weights, and strengthen the feature extraction capability. Finally, we built a dataset of waste classification supervision using the images provided by a city environmental protection bureau on which we did massive experiments. The experimental result shows that compared with other detection algorithms, the improved YOLO v3 algorithm has better performance. The mAP is 98.5%, which is 0.7% and 1.1% higher than the YOLOv5l and the EfficientDet-B0, respectively, and the average detection speed of the model is 14.6ms/it, which meets the requirements of regulatory real-time and environment complexity of the supervision system.

Eye Pupil Localisation and Labeling Using a Small Size Database and YOLOv4 Object Detection Algorithm

Eye-related research has shown that eye gaze data are very important for applications that are essential to human daily life. Eye gaze data has been used in research and systems for eye movements, eye tracking and eye gaze tracking. Eye pupil localization, labelling and tracking are challenging problems in computer science. This article discusses and explores that problem․ The YOLOv4 (“You only look once”) object detection algorithm which is an evolution of the YOLOv3 model has been evaluated in a tiny database consisting of 103 eye images. The YOLOv4 algorithm was created by Alexey Bochkovskiy, Chien-Yao Wang and Hong-Yuan Mark Liao [3]. It is twice as fast as EfficientDet with comparable performance. The main purpose of this article is to test the YOLOv4 algorithm, to find out its effectiveness in process of localization and labelling of eye pupils and find out (determine) the effectiveness of the algorithm when training with a tiny database and with a relatively small number of iterations.

High-efficiency tea shoot detection method via a compressed deep learning model

Achieving high-efficiency and accurate detection of tea shoots in fields is essential for tea robotic plucking. A real-time tea shoot detection method using the channel and layer pruned YOLOv3-SPP deep learning algorithm was proposed in this study. First, tea shoot images were collected and data augmentation was performed to increase sample diversity, and then a spatial pyramid pooling module was added to the YOLOv3 model to detect tea shoots. To simplify the tea shoot detection model and improve the detection speed, the channel pruning algorithm and the layer pruning algorithm were used to compress the model. Finally, the model was fine-tuned to restore its accuracy, and achieve the fast and accurate detection of tea shoots. The test results demonstrated that the number of parameters, model size, and inference time of the tea shoot detection model after compression reduced by 96.82%, 96.81%, and 59.62%, respectively, whereas the mean average precision of the model was only 0.40% lower than that of the original model. In the field test, the compressed model was deployed on a Jetson Xavier NX to conduct the detection of tea shoots. The experimental results demonstrated that the detection speed of the compressed model was 15.9 fps, which was 3.18 times that of the original model. All the results indicate that the proposed method could be deployed on tea harvesting robots with low computing power to achieve high efficiency and accurate detection. Keywords: deep learning, tea shoot detection, model compression, high-efficiency DOI: 10.25165/j.ijabe.20221503.6896 Citation: Li Y T, He L Y, Jia J M, Chen J N, Lyu J, Wu C Y. High-efficiency tea shoot detection method via a compressed deep learning model. Int J Agric & Biol Eng, 2022; 15(3): 159–166.

Multi-Target Detection of Table Tennis Video Based on CNN and Yolov3

In order to explore the effect of convolutional neural network (CNN) on the detection of athletes and balls in table tennis, and to solve the problems of low accuracy and weak generalization ability of table tennis athletes training data, this paper analyzed videos of table tennis athletes. For the detection of multiple targets in the videos, including athletes and balls, we used Yolov3 as a deep learning framework, and CNN as an automatic detection method when processing images. We trained and test the video data to improve the stability and accuracy of target detection, through modifying its model on the basis of the Yolov3 model. Finally, we detect the movement trajectories of athletes and balls in table tennis videos stably, and the accuracy is above 0.8.

Fruits and Vegetable Diseases Recognition Using Convolutional Neural Networks

As they have nutritional, therapeutic, so values, plants were regarded as important and they’re the main source of humankind’s energy supply. Plant pathogens will affect its leaves at a certain time during crop cultivation, leading to substantial harm to crop productivity & economic selling price. In the agriculture industry, the identification of fungal diseases plays a vital role. However, it requires immense labor, greater planning time, and extensive knowledge of plant pathogens. Computerized approaches are developed and tested by different researchers to classify plant disease identification, and that in many cases they have also had important results several times. Therefore, the proposed study presents a new framework for the recognition of fruits and vegetable diseases. This work comprises of the two phases wherein the phase-I improved localization model is presented that comprises of the two different types of the deep learning models such as You Only Look Once (YOLO)v2 and Open Exchange Neural (ONNX) model. The localization model is constructed by the combination of the deep features that are extracted from the ONNX model and features learning has been done through the convolutional-05 layer and transferred as input to the YOLOv2 model. The localized images passed as input to classify the different types of plant diseases. The classification model is constructed by ensembling the deep features learning, where features are extracted dimension of from pre-trained Efficientnetb0 model and supplied to next 07 layers of the convolutional neural network such as 01 features input, 01 ReLU, 01 Batch-normalization, 02 fully-connected. The proposed model classifies the plant input images into associated labels with approximately 95% prediction scores that are far better as compared to current published work in this domain.

Personal Assistant Robot

Since the boom in science and technology, humans have been trying to invent machines that could reduce their efforts in day to day activities. In this paper, we develop a personal assistant robot that could pick up objects and return it to the user. The robot is controlled using an android application in mobile phones. The robot can listen to user’s command and then respond in the best way possible. The user can command the robot to move to given location, capture images and pick objects. The robot is equipped with ultrasonic sensor and web camera that helps it to move to different location effectively. It is also equipped with sleds that play important role in object picking process. The robot uses a tiny YOLOv3 model which is rigorously trained on several images of the object. There are some possible improvements that can be achieved which could help this robot to be used in several other fields as well.

Detection of early bruises on apples using hyperspectral imaging combining with YOLOv3 deep learning algorithm

AbstractThe bruise on apples is a common damage mainly caused during the picking and transportation stages, but its detection robustness has not been satisfactory using the traditional machine learning method, and the precision of bruise detection with short formation time and light degree of bruise is low, and it is easily affected by the variable and complex environment, especially the bright spots. In this paper, a long‐wave near infrared (LW‐NIR 930–2,548 nm) hyperspectral imaging system for acquiring reflectance images of bruised apples were developed in this study. Then the regions of interest (ROIs) of sound and bruised tissues were analyzed and three specific spectral regions were determined. The segmented principal component analysis was performed to select characteristic wavelengths from the three regions. In order to verify the superiority over original hyperspectral data onto deep learning algorithms, the gray‐scale images of characteristic wavelengths (dataset I) and their PC images (dataset II) were input into YOLOv3 network to develop a bruise detection model, respectively. The results showed that all 110 bruise spots of testing set were identified correctly with the YOLOv3 model of dataset I and achieved a F1‐sore of 100% and FPS of 68, while the YOLOv3 model of dataset II and traditional detection method could not eliminate the interference of bright spots and misidentified them as bruises. In short, the YOLOv3 model based on the selected characteristic wavelengths has an excellent potential for on‐line detection of apple bruises.Practical ApplicationsDetection of apple bruises is essential to ensure the fruit quality of the same batch and the income of the farmers. In the past, most bruises were detected using traditional machine learning algorithms combined with hyperspectral imaging technology. This requires a series of cumbersome preprocessing and is susceptible to uneven brightness distribution of reflected image. Compared with the traditional method, our proposed detection model using hyperspectral imaging combining with YOLOv3 deep learning algorithm could obtain better detection accuracy and robustness, and avoid the problem of false detection caused by the interference of bright spots. Simultaneously, the method based on neural network provides a new idea for future multispectral online detection technology.

Object Recognition for Remote Sensing Images from UAVs via Convolutional Neural Networks

In recent years, researches of remote sensing data have begun to develop towards mass information and diversification, and traditional analytical methods alone could not meet the needs of modern data processing. Deep learning methods have developed a lot in the field of computer vision which has provided a new target recognition method for remote sensing images. This paper concentrates on the application of convolutional neural networks(CNN) in object recognition of Unmanned Aerial Vehicle(UAV) remote sensing images: YOLOv3 and Faster R-CNN network model are built on computer and NVIDIA JETSON TX2, and NWPU VHR-10 dataset is used as training and test samples for simulation verification. The results show that the accuracy of remote sensing multiclass target recognition using the YOLOv3 network model is 73.8%, and the detection rate is 0.25s per image on average. The accuracy of using YOLOv3 network model in several network models is second only to Faster RCNN, but its real-time performance is better than one of Faster RCNN. The YOLOv3 network model is more appropriate for carrying on the UAVs.

Using Channel and Network Layer Pruning Based on Deep Learning for Real-Time Detection of Ginger Images

Consistent ginger shoot orientation helps to ensure consistent ginger emergence and meet shading requirements. YOLO v3 is used to recognize ginger images in response to the current ginger seeder’s difficulty in meeting the above agronomic problems. However, it is not suitable for direct application on edge computing devices due to its high computational cost. To make the network more compact and to address the problems of low detection accuracy and long inference time, this study proposes an improved YOLO v3 model, in which some redundant channels and network layers are pruned to achieve real-time determination of ginger shoots and seeds. The test results showed that the pruned model reduced its model size by 87.2% and improved the detection speed by 85%. Meanwhile, its mean average precision (mAP) reached 98.0% for ginger shoots and seeds, only 0.1% lower than the model before pruning. Moreover, after deploying the model to the Jetson Nano, the test results showed that its mAP was 97.94%, the recognition accuracy could reach 96.7%, and detection speed could reach 20 frames·s−1. The results showed that the proposed method was feasible for real-time and accurate detection of ginger images, providing a solid foundation for automatic and accurate ginger seeding.

Intelligent Defect Detection Method for Additive Manufactured Lattice Structures Based on a Modified YOLOv3 Model

Intelligent Defect Detection Method for Additive Manufactured Lattice Structures Based on a Modified YOLOv3 Model