YOLOv3 Model Research Articles

Detection of typical obstacles in orchards based on deep convolutional neural network

Precise identification of farmland obstacles is an important environmental-perception task for agricultural vehicles. The orchard environment is a complex and unstructured environment, it is difficult to detect obstacles accurately and effectively. Hence, an improved lightweight object detection method based on YOLOv3 is proposed to identify typical obstacles in orchards, such as humans, cement columns and utility poles. The lightweight MobileNetV2 network was used in the model to reduce the running time when extracting image features, and Gaussian model was introduced to improve the detection effect. By reconstructing the loss function in the original YOLOv3 model, the proposed model can predict the accuracy of the positioning box. Based on the orchard image dataset which contains typical obstacles (humans, cement columns and utility poles) in orchards, comparative experiments were conducted on the accuracy and speed between the proposed model, Faster-RCNN, SSD and the original YOLOv3 model. The test results showed that the proposed model outperforms other models in terms of accuracy and speed. For the total test sets, the F1-score and mean average precision (mAP) were 91.76% and 88.64%, respectively. The prediction time of an image with the size of 416 pixels × 416 pixels on the GPU was 13 ms. Therefore, with a low memory requirement, high recognition accuracy, and fast recognition speed, the method proposed in this study can effectively detect typical obstacles in orchards environment, providing a basis for intelligent orchard robots to avoid obstacles.

A Comparison of YOLO Models for License Plate Detection

License plate detection and recognition (LPDR) has become an increasingly popular field of study under image recognition. The ability to autonomously create a bounding box around a license plate using machine learning models for image recognition is an important component of intelligent transportation systems. The YOLO model has become a popular CNN model for image detection because of its speed and accuracy. This work focuses on the use of the YOLO model for object localization as applied to license plate detection. The research compares the accuracy and speed between the YOLOv3 and YOLOv4 models along with various configuration parameters and finds that the YOLOv4 model is most accurate while the YOLOv3 with reduced image resolution is the fastest in prediction time.

Detection of microalgae objects based on the Improved YOLOv3 model.

Microalgae play a major role in the invasion of alien organisms with ballast water as a carrier, and traditional ballast water detection methods have many limitations in identifying microalgae species. Therefore, this paper proposes a method to identify microalgae in ballast water based on an Improved YOLOv3 model. The method first used a lightweight network MobileNet instead of the Darknet-53 network as the backbone network of feature extraction in the original YOLOv3 model. Secondly, improved spatial pyramid pooling (SPP) is introduced to pool and concatenate the multi-scale regional features so as to reduce the position error when detecting small objects. Then, by considering the overlap area of the bounding box, central point distance and aspect ratio, the Complete IoU (CIoU) algorithm is used to optimize the loss function of the YOLOv3 model. Finally, the proposed method is experimentally compared with other latest methods on the established dataset. The experimental results demonstrated that under the same conditions, this Improved YOLOv3 model achieves an average accuracy of 98.90%, and the detection efficiency is 8.59% higher than that of the original YOLOv3 model and is better than the existing methods. The average time of this method to identify a single image is 0.086 s, and it has a good detection effect on the identification of microalgae species.

Ensemble Method of Deep Learning, Color Segmentation, and Image Transformation to Track, Localize, and Count Cotton Bolls Using a Moving Camera in Real-Time

HighlightsAn ensemble method using color segmentation, deep learning, and image transformation was developed.Experiments were conducted to compare the method with other state-of-the-art tracking algorithms.The optimized ensemble method to track bolls achieved 94.4% accuracy using weakly trained tiny YOLOv2 models.The method achieved 7.6 frames per second and outperformed five other tracking methods.Abstract. In robotic applications, good perception can be computationally costly and create undesirable latency before a control decision is initiated. Most of the methods available for object detection deep learning are either fast with low accuracy or slow with high accuracy. Fast and accurate methods are necessary to track and localize objects such as cotton bolls that may be visible or occluded by each other or not well illuminated. In this study, an ensemble of a deep learning method and other image processing techniques was used to detect cotton bolls in-field on defoliated plants. In each image, a trained deep learning method, the YOLOv2 model, was used to detect open cotton bolls, and color segmentation was applied to confirm if the bolls detected by the YOLOv2 model were actually white to avoid false positives. Boll tracking was performed by following the spatial movement of good features on the edges of the bolls using the Lucas-Kanade algorithm. An image transformation algorithm was applied to the next image in case the previously detected boll was lost to retrieve the information of the missing boll. Each tracked and localized boll was stored and counted to give the total number of bolls detected. In this study, detection accuracy was sacrificed for image processing speed by using the YOLOv2 model. Detection accuracy was improved by using an ensemble method that combined image color segmentation, optical flow, and image transformation. This method was compared to eight other open-source methods implemented in OpenCV. The ensemble method detected and counted bolls at a speed of 7.6 fps with an accuracy of 94.4% using the Jetson TX2 embedded system to process 1K resolution images, outperforming the other OpenCV methods in various measurements. Keywords: Boll counting, Cotton, Cotton harvesting, DarkFlow, Darknet, Deep learning, Machine vision, YOLOv2.

Quantitative analysis of blood cells from microscopic images using convolutional neural network.

Blood cell count provides relevant clinical information about different kinds of disorders. Any deviation in the number of blood cells implies the presence of infection, inflammation, edema, bleeding, and other blood-related issues. Current microscopic methods used for blood cell counting are very tedious and are highly prone to different sources of errors. Besides, these techniques do not provide full information related to blood cells like shape and size, which play important roles in the clinical investigation of serious blood-related diseases. In this paper, deep learning-based automatic classification and quantitative analysis of blood cells are proposed using the YOLOv2 model. The model was trained on 1560 images and 2703-labeled blood cells with different hyper-parameters. It was tested on 26 images containing 1454 red blood cells, 159 platelets, 3 basophils, 12 eosinophils, 24 lymphocytes, 13 monocytes, and 28 neutrophils. The network achieved detection and segmentation of blood cells with an average accuracy of 80.6% and a precision of 88.4%. Quantitative analysis of cells was done following classification, and mean accuracy of 92.96%, 91.96%, 88.736%, and 92.7% has been achieved in the measurement of area, aspect ratio, diameter, and counting of cells respectively.Graphical abstract Graphical abstract where the first picture shows the input image of blood cells seen under a compound light microscope. The second image shows the tools used like OpenCV to pre-process the image. The third image shows the convolutional neural network used to train and perform object detection. The 4th image shows the output of the network in the detection of blood cells. The last images indicate post-processing applied on the output image such as counting of each blood cells using the class label of each detection and quantification of morphological parameters like area, aspect ratio, and diameter of blood cells so that the final result provides the number of each blood cell types (seven) and morphological information providing valuable clinical information.

Using an Improved YOLOv4 Deep Learning Network for Accurate Detection of Whitefly and Thrips on Sticky Trap Images

HighlightsThe proposed method detected thrips and whitefly more accurately than previous methods.The proposed method demonstrated good robustness to illumination reflections and different pest densities.Small pest detection was improved by adding large-scale feature maps and more residual units to a shallow network.Machine vision and deep learning created an end-to-end model to detect small pests on sticky traps in field conditions.Abstract. Pest detection is the basis of precise control in vegetable greenhouses. To improve the detection accuracy and robustness for two common small pests (whitefly and thrips) in greenhouses, this study proposes a novel small object detection approach based on the YOLOv4 model. Yellow sticky trap (YST) images at the original resolution (2560 × 1920 pixels) were collected using pest monitoring equipment in a greenhouse. The images were then cropped and labeled to create sub-images (416 × 416 pixels) to construct an experimental dataset. The labeled images used in this study (900 training, 100 validation, and 200 test) are available for comparative studies. To enhance the model’s ability to detect small pests, the feature map at the 8-fold downsampling layer in the backbone network was merged with the feature map at the 4-fold downsampling layer to generate a new layer and output a feature map with a size of 104 × 104 pixels. Furthermore, the residual units in the first two residual blocks were enlarged by four times to extract more shallow image features and the location information of target pests to withstand image degradation in the field. The experimental results showed that the mean average precision (mAP) for detection of whitefly and thrips using the proposed approach was improved by 8.2% and 3.4% compared with the YOLOv3 and YOLOv4 models, respectively. The detection performance slightly decreased as the pest densities increased in the YST image, but the mAP value was still 92.7% in the high-density dataset, which indicates that the proposed model has good robustness over a range of pest densities. Compared with previous similar studies, the proposed method has better potential to monitor whitefly and thrips using YSTs in field conditions. Keywords: Deep learning, Greenhouse pest management, Image processing, Pest detection, Small object, YOLOv4.

WITHDRAWN: People detection and counting using YOLOv3 and SSD models

Abstract Object detection has become a crucial task for the various applications used in the real world such as surveillance, security, and automated vehicle system. The counting of the numbers of peoples at any junction also having various applications to provide integrity to any task. To count the number of peoples at any junction, we have various methods. Among the present methods, we analyzed the two algorithms that are You Only Look Once (YOLOv3) and Single Shot multi-box Detector (SSD). Two tasks were performed independently; one is for object detection by using the image dataset, and the other one is the counting of objects by using the video dataset. In this research, these two methods are analyzed for counting as well as detection efficiency, and comparison is presented. The results have shown that the precision, recall, and F1 measure achieved for SSD is higher than YOLOV3 v3.

A novel data augmentation approach for mask detection using deep transfer learning

At the onset of 2020, the world saw the rise and spread of a global pandemic named COVID-19 which caused numerous deaths and affected millions of people around the world. Due to its highly contagious nature, this disease spread across the world within a short span of time. It forced almost all the nations to implement strict social distancing rules along with use of face masks to reduce the risk of getting infected. While the virus is still on loose, markets and business firms have reopened to keep the economy alive. This calls for modification of existing technological models to cater for the safety of individuals and stop the spread of virus in public places. One such stringent implementation to achieve this safety would be deployment of a mask detection model. The proposed mask detection models can serve as a vital utility in the coming years for ensuring proper enforcement of safety protocols. This research paper explores the use of state of the art YOLOv3 model, a deep transfer learning object detection technique, to develop a mask detection model. Along with the implementation of a standard approach of any object detection algorithm, this paper has proposed the use of a data augmentation approach for mask detection. The proposed model focuses on generating an augmented dataset from the standard dataset with the help of data augmentation done by using image filtering techniques such as grayscale and Gaussian blur. This augmented dataset is used for training the object detection model for mask detection. The mean average precision for the Data augmentation based mask detection model is observed to be 99.8% while training. Finally, a comparison on the model performance is evaluated for the standard and proposed augmented data approach. The experiment conducted showed that the average confidence level for Standard mask detection model was 0.94, 0.93, 0.91 for images of individuals (type A), images with groups of people (type B) and video with the group of people (type C) respectively. The average confidence levels for the Data augmentation based mask detection model for types A, B and C are 0.97, 0.96 0.93 respectively. This paper therefore concludes that the proposed Data augmentation based mask detection model performs better than the Standard mask detection model.

An Improved YOLOv3 Model for Asian Food Image Recognition and Detection

The detection and recognition of food pictures has become an emerging application field of computer vision. However, due to the small differences between the categories of food pictures and the large differences within the categories, there are problems such as missed inspections and false inspections in the detection and recognition process. Aiming at the existing problems, an improved YOLOv3 model of Asian food detection method is proposed. Firstly, increase the top-down fusion path to form a circular fusion, making full use of shallow and deep features. Secondly, introduce the convolution residual module to replace the ordinary convolution layer to increase the gradient correlation and non-linearity of the network. Thirdly, introduce the CBAM (Convolutional Block Attention Module) attention mechanism to improve the network’s ability to extract effective features. Finally, CIOU (Complete-IoU) loss is used to improve the convergence efficiency of the model. Experimental results show that the proposed improved model achieves better detection results on the Asian food UECFOOD100 data set.

A New Few-Shot Learning Method of Digital PCR Image Detection

With the global pandemic of infectious diseases, the demand for accurate nucleic acid detection is daily increasing. The traditional threshold-based algorithms are adopted as the mainstream for processing the images of digital polymerase chain reaction (dPCR) now, but they are facing huge challenges on complex problems such as irregular noise, uneven illumination, and the lack of data. So, this paper proposed a novel few-shot learning method based on our improved YOLOv3 model with fast processing speed and high accuracy to deal with complicated situations. Besides, to reduce the requirement of the large training dataset and annotation time of deep neural networks, we proposed the Random Background Transfer Method (RBTM) and Source Traceability Annotation Method (STAM) as the data augmentation and annotation method separately, which exploit the prior knowledge of the data and successfully realized the few-shot learning. Bases on the domain knowledge of dPCR images, our method could effectively augment images and reduce the labeling time by 70% while retaining the visually prominent features and improves the detection accuracy from 63.96% of the traditional threshold-based algorithm to as high as 98.98%. With the optimal processing speed and accuracy, our method is the state-of-art strategy for the detection of dPCR images now.

Monitoring the behaviours of pet cat based on YOLO model and raspberry Pi

ABSTRACT With the progress of the times and the rapid development of science and technology, machine learning and artificial intelligence are increasingly used in transportation, logistics, and homes. In terms of pets, pet monitoring has also become very popular in recent years. In this study, a real-time monitoring system for home pets using raspberry pie is developed. The proposed method consists of a raspberry Pi based YOLOv3-Tiny identification system for rapid detection and better boundary frame prediction of the cat behavior. Based on the YOLOv3-Tiny method, the following fine-tuning is implemented: (1) The images collected using the mobile phones are uniformly cropped to 416*416 pixels. (2) The images were captured in different periods and randomly rotated to plus or minus 20 degrees to make the dataset more robust for training. (3) One thousand four hundred pictures of the cat's movements in the room for marking and training are used. The dataset is also used to train the YOLOv3 model. According to the input image categories, the results of the output are categorized into six cat actions. They were sleeping, eating, sitting down, walking, going to the toilet, and search for a trash can. The average accuracy of both models was 98%. Based on environmental constraints like speed and memory consumption, any one of these models can be used for image recognition. The raspberry Pi system continuously captures the images and instantly sends a message to the registered mobile phone to achieve an instant preventive measure if the cat goes to the toilet for too long or flips through the trash can.

An Edge Computing‐Enabled Train Obstacle Detection Method Based on YOLOv3

Rail transit is developing towards intelligence which takes lots of computation resource to perform deep learning tasks. Among these tasks, object detection is the most widely used, like track obstacle detection, catenary wear, and defect detection and looseness detection of train wheel bolts. But the limited computation capability of the train onboard equipment prevents running deep and complex detection networks. The limited computation capability of the train onboard equipment prevents conducting complex deep learning tasks. Cloud computing is widely utilized to make up for the insufficient onboard computation capability. However, the traditional cloud computing architecture will bring in uncertain heavy traffic load and cause high transmission delay, which makes it fail to complete real‐time computing intensive tasks. As an extension of cloud computing, edge computing (EC) can reduce the pressure of cloud nodes by offloading workloads to edge nodes. In this paper, we propose an edge computing‐based method. The onboard equipment on a fast‐moving train is responsible for acquiring real‐time images and completing a small part of the inference task. Edge computing is used to help execute the object detection algorithm on the trackside and carry most of the computing power. YOLOv3 is selected as the object detection model, since it can balance between the real‐time and accurate performance on object detection compared with two‐stage models. To save onboard equipment computation resources and realize the edge‐train cooperative interface, we propose a model segmentation method based on the existing YOLOv3 model. We implement the cooperative inference scheme in real experiments and find that the proposed EC‐based object detection method can accomplish real‐time object detection tasks with little onboard computation resources.

InsuDet: A Fault Detection Method for Insulators of Overhead Transmission Lines Using Convolutional Neural Networks

One of the key tasks of the overhead line power equipment inspection based on aerial images acquired by unmanned aerial vehicles is to determine whether the insulators are faulty. However, the fault area on the insulator string occupies a relatively small portion of the entire image, which will make detection difficult. This article presents an intelligent fault detection method for overhead line insulators based on aerial images and improved you only look once (YOLOv3) deep learning technology. In our model, a densely connected feature pyramid network (FPN) is proposed. First, this network can improve the utilization rate of the strong semantic information of deep features and the localization information of shallow features, thereby improving the small insulator fault (missing-cap) detection performance of the YOLOv3 model. Second, this network reduces the number of parameters of the YOLOv3 model, resulting in a low risk of network over-fitting for small datasets. The experimental results on the CPLID dataset show that our model has higher detection accuracy in localization of overhead line insulators and detection of insulator missing-cap faults compared with the existing works.

Research on Automatic Recognition of Casting Defects Based on Deep Learning

A method for recognition of casting defects based on improved You Only Look Once (YOLO v3) is proposed to address the problems of slow detection speed, low detection efficiency, and poor robustness suffered from the current inspecting manually methods, which can improve the ability to detect defects, especially for tiny defects. Firstly, for obtained the industrial digital radiography images (DR images), we introduce the guide filtering technique to enhance the defects in these DR images, thus obtaining standard defect samples; Further, the defect samples are annotated to generate the defect detection data set for network training. In this article, the improved YOLOv3 network model structure is used to detect defects. Comparative experiments illustrate the proposed defect detection model for castings achieves better performance. Concretely, the experimental results show that the improved network model (YOLOv3_134) converges faster than the YOLOv3 network model and has better convergence than the YOLOv3 model. And the mean average precision (mAP) of the YOLOv3_134 is 26.1% higher than that of the original YOLOv3, which makes the YOLOv3_134 model-based casting defect detection method meet the industrial production requirements in terms of accuracy and speed.

Real-Time Object Detection Model

Object Detection is a study in the field of computer vision. An object detection model recognizes objects of the real world present either in a captured image or in real-time video where the object can belong to any class of objects namely humans, animals, objects, etc. This project is an implementation of an algorithm based on object detection called You Only Look Once (YOLO v3). The architecture of yolo model is extremely fast compared to all previous methods. Yolov3 model executes a single neural network to the given image and then divides the image into predetermined bounding boxes. These boxes are weighted by the predicted probabilities. After non max-suppression it gives the result of recognized objects together with bounding boxes. Yolo trains and directly executes object detection on full images.

Design and Development of a Smart Variable Rate Sprayer Using Deep Learning

The uniform application (UA) of agrochemicals results in the over-application of harmful chemicals, increases crop input costs, and deteriorates the environment when compared with variable rate application (VA). A smart variable rate sprayer (SVRS) was designed, developed, and tested using deep learning (DL) for VA application of agrochemicals. Real-time testing of the SVRS took place for detecting and spraying and/or skipping lambsquarters weed and early blight infected and healthy potato plants. About 24,000 images were collected from potato fields in Prince Edward Island and New Brunswick under varying sunny, cloudy, and partly cloudy conditions and processed/trained using YOLOv3 and tiny-YOLOv3 models. Due to faster performance, the tiny-YOLOv3 was chosen to deploy in SVRS. A laboratory experiment was designed under factorial arrangements, where the two spraying techniques (UA and VA) and the three weather conditions (cloudy, partly cloudy, and sunny) were the two independent variables with spray volume consumption as a response variable. The experimental treatments had six repetitions in a 2 × 3 factorial design. Results of the two-way ANOVA showed a significant effect of spraying application techniques on volume consumption of spraying liquid (p-value < 0.05). There was no significant effect of weather conditions and interactions between the two independent variables on volume consumption during weeds and simulated diseased plant detection experiments (p-value > 0.05). The SVRS was able to save 42 and 43% spraying liquid during weeds and simulated diseased plant detection experiments, respectively. Water sensitive papers’ analysis showed the applicability of SVRS for VA with >40% savings of spraying liquid by SVRS when compared with UA. Field applications of this technique would reduce the crop input costs and the environmental risks in conditions (weed and disease) like experimental testing.

SSS-YOLO: towards more accurate detection for small ships in SAR image

ABSTRACT Aiming at the low detection rate and high false alarm in small ship detection in SAR images, we propose a small-scale ship detection algorithm based on convolutional neural network in this paper. First, we redesign the feature extraction network according to the characters of ship targets in SAR images. The modified network can enrich the spatial and semantics information of small ships. Then, we propose the Path Argumentation Fusion Network (PAFN) to improve the fusion of different feature maps. PAFN uses bottom-up and top-down ways to fuse more location information and semantic information. Both these two optimizations can enhance the detection for small ships. We evaluate our model based on the open SAR-Ship-Dataset and Gaofen-3 SAR images. The experiment results show that our method has excellent performance for small ship detection compared with other deep learning models. Our model improves AP by 6.5% and has higher detection efficiency compared with the baseline YOLOv3 model.

Intelligent Traffic Congestion Control System using Machine Learning and Wireless Network

Traffic congestion has become a big problem for most people because it increases noise, air pollution, and wasting time. Current normal traffic light system is not enough to manage the traffic problematic congestions because they operate on a fixed-time length plan. In recent years, internet of things led to introducing new models of intelligent traffic light systems; by utilizing different techniques such as predictive-based model, radiofrequency identification, and ultrasonic-based model. The most essential one of these techniques is depends of image processing and microcontroller communications. In this paper, we propose an intelligent, low cost, and efficient microcontroller circuit-based system for controlling cars in traffic light. This system can manage car traffics smarter than traditional approaches, it is capable to dynamically adjust timings of traffic signal. It can rapidly respond to traffic conditions to reduce traffic congestion. For implementing this system, a server, microcontroller board, cameras, as hardware and wireless network between traffic lights as infrastructure for communication are used. The system uses machine learning technique (i.e.,Yolov3 model and OpenCV) for decision depending on existence of emergency cars and number of cars. The experiment results show higher accuracy in managing traffic lights and recognizing the emergency cars.

Vehicle Detection Using Deep Learning Technique in Tunnel Road Environments

This paper proposes a real-time detection method for a car driving ahead in real time on a tunnel road. Unlike the general road environment, the tunnel environment is irregular and has significantly lower illumination, including tunnel lighting and light reflected from driving vehicles. The environmental restrictions are large owing to pollution by vehicle exhaust gas. In the proposed method, a real-time detection method is used for vehicles in tunnel images learned in advance using deep learning techniques. To detect the vehicle region in the tunnel environment, brightness smoothing and noise removal processes are carried out. The vehicle region is learned after generating a learning image using the ground-truth method. The YOLO v2 model, with an optimal performance compared to the performances of deep learning algorithms, is applied. The training parameters are refined through experiments. The vehicle detection rate is approximately 87%, while the detection accuracy is approximately 94% for the proposed method applied to various tunnel road environments.

An Efficient Small Traffic Sign Detection Method Based on YOLOv3

In recent years, target detection framework based on deep learning has made brilliant achievements. However, real-life traffic sign detection remains a great challenge for most of the state-of-the-art object detection methods. The existing deep learning models are inadequate to effectively extract the features of small traffic signs from large images in real-world conditions. In this paper, we address the small traffic sign detection challenge by proposing a novel small traffic sign detection method based on a highly efficient end-to-end deep network model. The proposed method features fast speed and high precision as it attaches three key insights to the established You Only Look Once (YOLOv3) architecture and other correlated algorithms. Besides, network pruning is appropriately introduced to minimize network redundancy and model size while keeping a competitive detection accuracy. Furtherly, four scale prediction branches are also adopted to significantly enrich the feature maps of multi-scales prediction. In our method, we adjust the loss function to balance the contribution of error source to the total loss. The effectiveness, and robustness of the network is further proved with experiments on Tsinghua-Tencent 100 K traffic sign dataset. The experimental results indicate that our proposed method has achieved better accuracy than that of the original YOLOv3 model. Compared with the schemes in relevant literatures our proposed method not only emerges performance superiors in detection recall and accuracy, but also achieves 1.9–2.7x improvement in detection speed.