Deep Learning Object Detection Research Articles

Design of apple recognition model based on improved deep learning object detection framework Faster-RCNN

It the stage of digital economy development in China, intelligent recognition technology is used in agriculture, forestry and planting industries. This paper improves and optimizes apple recognition based on Faster-RCNN, a deep learning target detection framework, and analyzes the advantages and disadvantages of the improved target detection and recognition model. This study compares and analyzes the conventional Faster-RCNN apple recognition model and the enhanced Faster-RCNN apple recognition model design based on the deep learning target detection framework. From the two groups of recognition model systems under different design schemes, the research and analysis of comprehensive performance, sensitivity and coupling shows that the improved Faster-RCNN Apple recognition model has higher accuracy and more accurate positioning for Apple recognition. The results show that the improved Faster-RCNN apple recognition model can more effectively improve the design of the apple recognition model at the present stage, reduce labor costs, and significantly improve work efficiency. It also promotes the development of the domestic digital economy and provides research support and value for future intelligence and deep learning development paths.

Investigating the Sim-to-Real Generalizability of Deep Learning Object Detection Models.

State-of-the-art object detection models need large and diverse datasets for training. As these are hard to acquire for many practical applications, training images from simulation environments gain more and more attention. A problem arises as deep learning models trained on simulation images usually have problems generalizing to real-world images shown by a sharp performance drop. Definite reasons and influences for this performance drop are not yet found. While previous work mostly investigated the influence of the data as well as the use of domain adaptation, this work provides a novel perspective by investigating the influence of the object detection model itself. Against this background, first, a corresponding measure called sim-to-real generalizability is defined, comprising the capability of an object detection model to generalize from simulation training images to real-world evaluation images. Second, 12 different deep learning-based object detection models are trained and their sim-to-real generalizability is evaluated. The models are trained with a variation of hyperparameters resulting in a total of 144 trained and evaluated versions. The results show a clear influence of the feature extractor and offer further insights and correlations. They open up future research on investigating influences on the sim-to-real generalizability of deep learning-based object detection models as well as on developing feature extractors that have better sim-to-real generalizability capabilities.

Targeted weed management of Palmer amaranth using robotics and deep learning (YOLOv7).

Effective weed management is a significant challenge in agronomic crops which necessitates innovative solutions to reduce negative environmental impacts and minimize crop damage. Traditional methods often rely on indiscriminate herbicide application, which lacks precision and sustainability. To address this critical need, this study demonstrated an AI-enabled robotic system, Weeding robot, designed for targeted weed management. Palmer amaranth (Amaranthus palmeri S. Watson) was selected as it is the most troublesome weed in Nebraska. We developed the full stack (vision, hardware, software, robotic platform, and AI model) for precision spraying using YOLOv7, a state-of-the-art object detection deep learning technique. The Weeding robot achieved an average of 60.4% precision and 62% recall in real-time weed identification and spot spraying with the developed gantry-based sprayer system. The Weeding robot successfully identified Palmer amaranth across diverse growth stages in controlled outdoor conditions. This study demonstrates the potential of AI-enabled robotic systems for targeted weed management, offering a more precise and sustainable alternative to traditional herbicide application methods.

The Potential of Deep Learning Object Detection in Citizen-Driven Snail Host Monitoring to Map Putative Disease Transmission Sites

The Potential of Deep Learning Object Detection in Citizen-Driven Snail Host Monitoring to Map Putative Disease Transmission Sites

A Dataset of Visible Light and Thermal Infrared Images for Health Monitoring of Caged Laying Hens in Large-Scale Farming.

Considering animal welfare, the free-range laying hen farming model is increasingly gaining attention. However, in some countries, large-scale farming still relies on the cage-rearing model, making the focus on the welfare of caged laying hens equally important. To evaluate the health status of caged laying hens, a dataset comprising visible light and thermal infrared images was established for analyses, including morphological, thermographic, comb, and behavioral assessments, enabling a comprehensive evaluation of the hens' health, behavior, and population counts. To address the issue of insufficient data samples in the health detection process for individual and group hens, a dataset named BClayinghens was constructed containing 61,133 images of visible light and thermal infrared images. The BClayinghens dataset was completed using three types of devices: smartphones, visible light cameras, and infrared thermal cameras. All thermal infrared images correspond to visible light images and have achieved positional alignment through coordinate correction. Additionally, the visible light images were annotated with chicken head labels, obtaining 63,693 chicken head labels, which can be directly used for training deep learning models for chicken head object detection and combined with corresponding thermal infrared data to analyze the temperature of the chicken heads. To enable the constructed deep-learning object detection and recognition models to adapt to different breeding environments, various data enhancement methods such as rotation, shearing, color enhancement, and noise addition were used for image processing. The BClayinghens dataset is important for applying visible light images and corresponding thermal infrared images in the health detection, behavioral analysis, and counting of caged laying hens under large-scale farming.

YOLO-CIANNA: Galaxy detection with deep learning in radio data

Context. The upcoming Square Kilometer Array (SKA) will set a new standard regarding data volume generated by an astronomical instrument, which is likely to challenge widely adopted data-analysis tools that scale inadequately with the data size. Aims. The aim of this study is to develop a new source detection and characterization method for massive radio astronomical datasets based on modern deep-learning object detection techniques. For this, we seek to identify the specific strengths and weaknesses of this type of approach when applied to astronomical data. Methods. We introduce YOLO-CIANNA, a highly customized deep-learning object detector designed specifically for astronomical datasets. In this paper, we present the method and describe all the elements introduced to address the specific challenges of radio astronomical images. We then demonstrate the capabilities of this method by applying it to simulated 2D continuum images from the SKA observatory Science Data Challenge 1 (SDC1) dataset. Results. Using the SDC1 metric, we improve the challenge-winning score by +139% and the score of the only other post-challenge participation by +61%. Our catalog has a detection purity of 94% while detecting 40–60% more sources than previous top-score results, and exhibits strong characterization accuracy. The trained model can also be forced to reach 99% purity in post-process and still detect 10–30% more sources than the other top-score methods. It is also computationally efficient, with a peak prediction speed of 500 images of 512×512 pixels per second on a single GPU. Conclusions. YOLO-CIANNA achieves state-of-the-art detection and characterization results on the simulated SDC1 dataset and is expected to transfer well to observational data from SKA precursors.

Infield disease detection in citrus plants: integrating semantic segmentation and dynamic deep learning object detection model for enhanced agricultural yield

Infield disease detection in citrus plants: integrating semantic segmentation and dynamic deep learning object detection model for enhanced agricultural yield

Comparison of Faster R-CNN, YOLO, and SSD for Third Molar Angle Detection in Dental Panoramic X-rays.

The use of artificial intelligence algorithms (AI) has gained importance for dental applications in recent years. Analyzing AI information from different sensor data such as images or panoramic radiographs (panoramic X-rays) can help to improve medical decisions and achieve early diagnosis of different dental pathologies. In particular, the use of deep learning (DL) techniques based on convolutional neural networks (CNNs) has obtained promising results in dental applications based on images, in which approaches based on classification, detection, and segmentation are being studied with growing interest. However, there are still several challenges to be tackled, such as the data quality and quantity, the variability among categories, and the analysis of the possible bias and variance associated with each dataset distribution. This study aims to compare the performance of three deep learning object detection models-Faster R-CNN, YOLO V2, and SSD-using different ResNet architectures (ResNet-18, ResNet-50, and ResNet-101) as feature extractors for detecting and classifying third molar angles in panoramic X-rays according to Winter's classification criterion. Each object detection architecture was trained, calibrated, validated, and tested with three different feature extraction CNNs which are ResNet-18, ResNet-50, and ResNet-101, which were the networks that best fit our dataset distribution. Based on such detection networks, we detect four different categories of angles in third molars using panoramic X-rays by using Winter's classification criterion. This criterion characterizes the third molar's position relative to the second molar's longitudinal axis. The detected categories for the third molars are distoangular, vertical, mesioangular, and horizontal. For training, we used a total of 644 panoramic X-rays. The results obtained in the testing dataset reached up to 99% mean average accuracy performance, demonstrating the YOLOV2 obtained higher effectiveness in solving the third molar angle detection problem. These results demonstrate that the use of CNNs for object detection in panoramic radiographs represents a promising solution in dental applications.

HELMET DETECTION AND LICENSE PLATE RECOGNITION USING CNN

Right now, we're facing a lot of issues with traffic rules in India that might use some fresh perspectives to fix. There has been a rise in the number of accidents and fatalities in India caused by the traffic offense of riding motorcycles or mopeds without a helmet. The current system mostly relies on CCTV records to track traffic offenses. In such cases, the traffic police have to zoom in on the license plate to identify the offending rider if they aren't wearing a helmet. The traffic offenses are common, and the number of persons riding motorbikes is growing daily, so this takes a lot of time and effort. Imagine a system that could detect whether a motorcyclist or moped rider isn't wearing a helmet and, if found, immediately get the license plate number. Recent studies have effectively accomplished this task using various characteristics such as CNN, R-CNN, LBP, HoG, HaaR, etc. In terms of speed, accuracy, and efficiency, however, these works have their limitations when it comes to object recognition and categorization. To try to automate the process of finding drivers who don't wear helmets and getting their license plate numbers, this study developed a system called Non-Helmet Rider detection. The core idea is based on three-level deep learning for object detection. Using YOLOv2, the first level detects a person and a motorbike or moped; the second level uses YOLOv3, and the third level uses YOLOv2. The items recognized are a helmet and a license plate. After then, Optical Character Recognition is used to obtain the license plate registration number. There are some limitations and requirements placed on all of these methods, particularly the one that extracts license plates. The efficiency of the process is crucial since video is used as an input in this task. We have developed a comprehensive system that can recognize helmets and retrieve license plate numbers using the aforementioned approaches.

Identifying sulphurous water discharge from legacy oil and gas wells using spectral band analysis of aerial and satellite imagery

Legacy oil and gas wells are a significant source of hydrogen sulphide and methane gas release to the atmosphere. Unanticipated occurrences of gas release in urbanized areas can pose human health risks. Several high hydrogen sulphide-emitting wells have been identified in Norfolk County, Ontario, Canada, based on the remote detection of sulphurous water discharging to the surface along the wellbore. Although oil and gas well records report well status and location, community reports of noxious hydrogen sulphide odours suggest potentially compromised well seals and inaccuracies in their documented location. Given the broad-scale nature of this issue and limitations in site access, a remote-sensing based methodology utilizing satellite and high-resolution aerial photography could support identification of undocumented sulphurous water leaks associated with compromised oil and gas wells and subsequent characterization of hydrological and ecohydrological impacts over time. This study presents a multifaceted approach to identifying sulphurous leaking wells utilizing complimentary remote sensing imagery and image analysis tools within ArcGIS. Southwestern Ontario Orthophotography Project air photos and Sentinel-2 satellite imagery were determined to be the most applicable sources of imagery based on their respective advantages in resolution and revisit time. The application of a normalized difference vegetation index showed that major well leaks could exhibit signs of vegetative scarcity, while minor well leaks may have only a limited impact on vegetation. A band combination utilizing the green and near infrared bands was created to enhance the detection of sulphurous water leaks. Utilizing the identified band combination, a pair of potentially undiscovered leaks were located west of a fluvial river channel. Continued research should attempt to employ an automated approach for discerning additional sulphurous water leaks in the region or at different points in time. Possible techniques may involve the deep learning object and change detection models incorporated in ArcGIS.

Facemask detection and classification using you only look once version 7

<p>World Health Organization (WHO) suggests that wearing masks and keeping social distancing are the best ways to avoid infection transmission of communicable diseases. Consequently, most governments have forced people to wear masks in public areas to prevent communicable diseases such as COVID-19. Manual monitoring and surveillance are time-consuming and not always possible in crowded areas. Hence, object detection deep learning models can effectively handle these challenges. Therefore, this work aims to investigate the efficiency of different versions of the you only look once version 7 (YOLOv7) model in facemask detection and classification over the privately balanced dataset. The dataset comprises of 1,300 images with four novel classes; including no occlusion, correct mask, incorrect mask, and other use cases. Furthermore, the model’s performance was evaluated based on mean average precision (mAP), recall, precision, and inference time. Finally, a comparative result analysis has been reported to determine the best model for facemask detection and classification. YOLOv7 model versions exhibit widely various performances ranging from 20.7% mAP for YOLOv7-D6 to 95.5% for YOLOv7-tiny. In contrast, the inference time for all YOLOv7 versions covers a narrow range of 3 ms. In conclusion, the YOLOv7-tiny version outperforms other models, achieving a high detection performance and acceptable detection speed.</p>

Mangrove species detection using YOLOv5 with RGB imagery from consumer unmanned aerial vehicles (UAVs)

Despite comprising only one per cent of global forests, mangroves provide vital ecological and economic benefits to their ecosystems. Due to its decreasing extent over the past decade, there is a rise in research innovations supporting mangrove conservation. Specifically, consumer-grade Unmanned Aerial Vehicles (UAV) were proven effective as potential remote sensing alternatives to support mangrove research and monitoring in recent studies. As most studies use custom UAV-mounted sensors for mangrove species classification, similar studies using a UAV’s default red–green–blue (RGB) cameras were scarce. This study explores the potential of high-resolution RGB aerial images through state-of-the-art object detection algorithm, YOLOv5 to detect the dominant Rhizophora mangroves in Sarawak, Malaysia. A total of 400 RGB images were equally selected from two study areas and allocated into three datasets, two corresponding to each study area and one combining all images. The annotation process was performed using a previously proposed novel method, assisted by YOLOv5 for a semi-automated annotation process with expert verification. Systematic training experiments were conducted to select an optimal epoch size across models trained with each dataset. The final models produced an average true positive rate of 73.8% and 71.7% for each study site, while the combined dataset model produced an average true positive rate of 73.7%. Overall, this study demonstrated the potential of UAV-based RGB images and deep learning object detection architectures to identify specific mangrove objects, while also highlighting key considerations for similar future research.

A Lightweight Crop Pest Detection Method Based on Improved RTMDet

To address the issues of low detection accuracy and large model parameters in crop pest detection in natural scenes, this study improves the deep learning object detection model and proposes a lightweight and accurate method RTMDet++ for crop pest detection. First, the real-time object detection network RTMDet is utilized to design the pest detection model. Then, the backbone and neck structures are pruned to reduce the number of parameters and computation. Subsequently, a shortcut connection module is added to the classification and regression branches, respectively, to enhance its feature learning capability, thereby improving its accuracy. Experimental results show that, compared to the original model RTMDet, the improved model RTMDet++ reduces the number of parameters by 15.5%, the computation by 25.0%, and improves the mean average precision by 0.3% on the crop pest dataset IP102. The improved model RTMDet++ achieves a mAP of 94.1%, a precision of 92.5%, and a recall of 92.7% with 4.117M parameters and 3.130G computations, outperforming other object detection methods. The proposed model RTMDet++ achieves higher performance with fewer parameters and computations, which can be applied to crop pest detection in practice and aids in pest control research.

Detection of kernels in maize forage using hyperspectral imaging

In this study, the use of hyperspectral imaging was examined to enhance the kernel processing assessment of maize forage, which is crucial for optimizing the production of dairy cow feed and biogas production. As the visual contrast between the kernels and other crop particles is quite limited, spectral imaging could provide better kernel classification. A pushbroom hyperspectral imaging system was used, scanning samples collected during maize harvesting in the 400–1000 nm range. A PLSDA model for pixel classification was developed to distinguish kernel spectra from the other particles in the forage, achieving a pixel-level classification accuracy of 95.2 %. Next, the most important wavelengths were identified by means of a stepwise procedure using the Wilks Lambda criterion. A Pixel classification using the top five discriminating wavelengths achieved nearly the same accuracy as the full spectrum wavelength model (95.2 % compared to 93.5 %) and did considerably better than the RGB classifier’s 86.3 % accuracy. Finally, these top 5 discriminating wavebands were applied in an object detection deep learning model, more specifically a Faster R-CNN model. While object detection based on these 5 wavebands stilled outperformed object detection based on RGB wavebands, detection performance, as measured by AP50, was rather low. This weak performance resulted from the low resolution of the hyperspectral imaging camera.

Field-based multispecies weed and crop detection using ground robots and advanced YOLO models: A data and model-centric approach

The implementation of a machine-vision system for real-time precision weed management is a crucial step towards the development of smart spraying robotic vehicles. The intelligent machine-vision system, constructed using deep learning object detection models, has the potential to accurately detect weeds and crops from images. Both data-centric and model-centric approaches of deep learning model development offer advantages depending on environment and non-environment factors. To assess the performance of weed detection in real-field conditions for eight crop species, the Yolov8, Yolov9, and customized Yolov9 deep learning models were trained and evaluated using RGB images from four locations (Casselton, Fargo, and Carrington) over a two-year period in the Great Plains region of the U.S. The experiment encompassed eight crop species—dry bean, canola, corn, field pea, flax, lentil, soybean, and sugar beet—and five weed species—horseweed, kochia, redroot pigweed, common ragweed, and water hemp. Six Yolov8 and eight Yolov9 model variants were trained using annotated weed and crop images gathered from four different sites including combined dataset from four sites. The Yolov8 and Yolov9 models’ performance in weed detection were assessed based on mean average precision (mAP50) metrics for five datasets, eight crop species, and five weed species. The results of the weed and crop detection evaluation showed high mAP50 values of 86.2 %. The mAP50 values for individual weed and crop species detection ranged from 80.8 % to 98 %. The results demonstrated that the performance of the model varied by model type (model-centric), location due to environment (data-centric), data size (data-centric), data quality (data-centric), and object size in the image (data-centric). Nevertheless, the Yolov9 customized lightweight model has the potential to play a significant role in building a real time machine-vision-based precision weed management system.

Sign Language Detection Using Machine Learning

This paper investigates the application of machine learning for sign language detection. The objective is to develop a model that translates sign language into spoken language, bridging the communication gap between deaf and hearing individuals. You Only Look Once (YOLO), a deep learning object detection algorithm, is employed to train a model on a dataset of labeled sign language images derived from video data. The system achieves real-time sign detection in videos. However, challenges include the scarcity of large, labeled datasets and the inherent ambiguity of certain signs, which can lead to reduced detection accuracy. This research contributes to the field of Assistive Technologies (AT) by promoting accessibility and social inclusion for the deaf community.

Optimizing the spatial accessibility of outdoor sports facilities: a greedy heuristic algorithm based on remote sensing images

ABSTRACT Recent studies on the accessibility of sports facilities have rarely considered the specific attributes of the facilities, limiting their ability to define service potential, and have often neglected the critical aspect of equitable access. This study proposed a novel approach based on remote sensing images to optimize the spatial accessibility of outdoor sports facilities. Using Shanghai, China, as the study area, the study identified four types of sports facilities using a deep learning object detection method, which allowed their service capacities (areas) to be measured more precisely. A greedy heuristic algorithm was then developed based on a "trade-off" strategy that seeks to optimize facility access by reconciling the objectives of enhancing access and ensuring equality and by weighing the benefits of utilizing existing resources (school facilities) against the necessity of developing new ones. The object detection method achieved precision and recall rates of 88% and 96%, respectively, and the optimization efforts resulted in a 73% increase in accessibility while also significantly reducing the Gini coefficient from 0.58 to 0.34. The proposed algorithm outperformed the random selection and all-school-opening strategies. The results indicated that the methodology can effectively create refined datasets for outdoor sports facilities and enhance their accessibility.

Explicit Incorporation of Spatial Autocorrelation in 3D Deep Learning for Geospatial Object Detection

Three-dimensional (3D) geospatial object detection has become essential for 3D geospatial studies driven by explosive growth in 3D data. It is extremely labor- and cost-intensive, though, as it often requires manual detection. Deep learning has been recently used to automate object detection within 3D context. Yet, addressing spatial dependency in 3D data and how it might inform deep learning for 3D geospatial object detection remains a significant challenge. Traditional models focus on the use of spatial properties, often overlooking color and contextual information. Exploiting these nonspatial attributes for 3D geospatial object detection thus becomes crucial. Our study pioneers explicit incorporation of spatial autocorrelation of color information into 3D deep learning for object detection. We introduce an innovative framework to estimate spatial autocorrelation, addressing challenges in unstructured 3D data sets. Our experiments suggest the effectiveness of incorporating spatial autocorrelation features in enhancing the accuracy of 3D deep learning models for geospatial object detection. We further investigate the uncertainty of such contextual information brought by diverse configurations, exemplified by the number of nearest neighbors. This study advances 3D geospatial object detection via using spatial autocorrelation to inform deep learning algorithms, strengthening the connection between GIScience and artificial intelligence and, thus, holding implications for diverse GeoAI applications.

Foreign Object Debris Detection on Wireless Electric Vehicle Charging Pad Using Machine Learning Approach

Foreign object debris (FOD) includes any unwanted and unintentional material lying on the charging lane or parking lots, posing a risk to the wireless charging system, the vehicle, or the people inside. FOD in an Electric Vehicle (EV) wireless charging system can cause problems, including decreased charging efficiency, safety risks, charging system damage, communication issues, and health risks. To address this problem, this paper proposes the deep learning object detection network approach of using YOLOv4 (You Only Look Once), which is a single-shot detector. Additionally, for real-time implementation, YOLOv4-Tiny is suggested, which is a compressed version of YOLOv4 designed for devices with low computational power. YOLOv4-Tiny enables faster inferences and facilitates the deployment of FOD detectors on edge devices. The algorithm is trained using the FOD dataset, consisting of images of common debris on runways or taxiways. Furthermore, utilizing the concept of transfer learning, the last few layers of the pre-trained YOLOv4 model are modified using the COCO (Common Objects in Context) dataset to transfer features to the new network and retrain the model on the FOD dataset. The results obtained using this YOLOv4 model yielded a precision rate of 99.05%, while the results from YOLOv4-Tiny achieved a precision rate of 97.74%, with an average inference time of 150 ms under the ambient light and weather conditions.

Lane and Object Detection using YOLO: Indian Roads Scenario

This research explores the feasibility of You Only Look Once (YOLO), a deep learning object detection algorithm, for lane and object detection in challenging Indian road environments. Traditional methods struggle with faded lane markings, dense and diverse traffic, and unpredictable scenarios. YOLO's speed and accuracy make it suitable for real-time ADAS applications. The methodology leverages Roboflow, a platform for computer vision tasks, to explore data acquisition, model selection, training, and evaluation. This research aims to contribute to developing safer and more reliable ADAS systems for Indian roads.