Digital framework for georeferenced multiplatform surveillance of banana wilt using human in the loop AI and YOLO foundation models

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 28, 2025

Bananas (Musa spp.) are a critical global food crop, providing primary source of nutrition for millions people. Traditional methods disease monitoring and detection often time-consuming, labor-intensive, prone to inaccuracies. This study introduces an AI-powered multiplatform georeferenced surveillance system designed enhance the management banana wilt diseases. We developed evaluated several deep learning foundation models, including YOLO-NAS, YOLOv8, YOLOv9, Faster-RCNN perform accurate on both platforms. Our results demonstrate superior performance YOLOv9 in detecting healthy, Fusarium Wilt Xanthomonas diseased plants aerial images, achieving high mAP@50, precision recall metrics ranging from 55 86%. In terms ground level we organized dataset based occurrence Africa, Latin America, India, Asia Australia. For this platform, YOLOv8 outperforms rest achieves between 65 99% depending plant part region. Additionally, incorporated Explainable AI techniques, such as Gradient-weighted Class Activation Mapping, model transparency trustworthiness. Human Loop Artificial Intelligence was also utilized model's predictions.

Language: Английский

---

YOLO-SIFD: YOLO with Sliced Inference and Fractal Dimension Analysis for Improved Fire and Smoke Detection

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2025, Volume and Issue: 82(3), P. 5343 - 5361

Published: Jan. 1, 2025

Language: Английский

---

Effectiveness of YOLO Architectures in Tree Detection: Impact of Hyperparameter Tuning and SGD, Adam, and AdamW Optimizers

Standards, Journal Year: 2025, Volume and Issue: 5(1), P. 9 - 9

Published: March 17, 2025

This study investigates the optimization of tree detection in static images using YOLOv5, YOLOv8, and YOLOv11 models, leveraging a custom non-standard image bank created exclusively for this research. Objectives: To enhance by comparing performance models. The comparison involved hyperparameter tuning application various optimizers, aiming to improve model terms precision, recall, F1, mean average precision (mAP). Design/Methodology/Approach: A was utilized train During training, hyperparameters’ learning rate momentum were tuned combination with optimizers SGD, Adam, AdamW. Performance metrics, including mAP, analyzed each configuration. Key Results: process achieved values 100% Adam YOLOv8 SGD YOLOv11, recall 91.5% AdamW on YOLOv8. Additionally, mAP reached 95.6% 95.2% YOLOv11. Convergence times also significantly reduced, demonstrating faster training enhanced overall performance. Originality/Research gap: addresses gap YOLO models trained banks, topic that is less commonly explored literature. exclusive development further adds novelty Practical Implications: findings underscore effectiveness tasks datasets. methodology could be extended other applications requiring object banks. Limitations investigation: limited within single dataset does not evaluate generalizability these optimizations datasets or tasks.

Language: Английский

---

Research and application of deep learning object detection methods for forest fire smoke recognition

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 10, 2025

Language: Английский

---

YOLOv8 Model Architecture Selection for Human Fall Detection

Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 219 - 227

Published: Jan. 1, 2025

Language: Английский

---

The Accelerated Inference of a Novel Optimized YOLOv5-LITE on Low-Power Devices for Railway Track Damage Detection

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 134846 - 134865

Published: Jan. 1, 2023

Railway track malfunctions can lead to severe consequences such as train derailments and collisions. Traditional manual inspection methods suffer from inaccuracies low efficiency. Contemporary deep learning-based detection techniques have challenges in model accuracy, inference speed, are often associated with expensive computational costs high power consumption when deployed on devices. We propose an optimized lightweight network based YOLOv5-lite. which employs enhanced Fused Mobile Inverted Bottleneck Convolution (BF_MBConv) reduce the number of parameters floating-point operations (FLOP) during backbone feature extraction. The Squeeze-and-Excitation (SE) mechanism is adopted, emphasizing more critical features by assigning different weights a channel-wise perspective. Utilizing DropBlock holistic dropping substitute for Dropout random offers efficient means discarding redundant features. In neck section, Shuffle convolution replaces conventional one, significantly reducing parameter count while better integrating information post-group convolution. Lastly, incorporation Focal-EIoU Loss augments regression, application incremental dataset processing techniques, it addresses accuracy sample imbalance issues. refined algorithm achieves mean Average Precision (mAP)@0.5 94.4%, marking 8.13% improvement over original Moreover, leveraging embedded platform integrated Intel® Movidius™ Neural Compute Stick cluster portable device deployment, Achieved frame rate 18.7 FPS. Our findings indicate that this approach efficiently accurately detect railway damages. Additionally, previously overlooked issues performance-cost trade-offs, countering past trend prioritizing performance at expense elevated costs, proposing harmonized prioritizes efficiency affordability.

Language: Английский

---

Multi-Dimensional Information Fusion You Only Look Once Network for Suspicious Object Detection in Millimeter Wave Images

Electronics, Journal Year: 2024, Volume and Issue: 13(4), P. 773 - 773

Published: Feb. 16, 2024

Millimeter wave (MMW) imaging systems have been widely used for security screening in public places due to their advantages of being able detect a variety suspicious objects, non-contact operation, and harmlessness the human body. In this study, we propose an innovative, multi-dimensional information fusion YOLO network that can aggregate capture multimodal cope with challenges low resolution susceptibility noise MMW images. particular, data aggregation module is developed adaptively synthesize novel type image, which simultaneously contains pixel, depth, phase, diverse signal-to-noise overcome limitations current images containing consistent pixel all three channels. Furthermore, capable differentiable enhancements take into account adverse conditions real application scenarios. order fully acquire augmented contextual mentioned above, asymptotic path combine it YOLOv8. The proposed method bidirectionally fuse deep shallow features while avoiding semantic gaps. addition, multi-view, multi-parameter mapping technique designed enhance detection ability. experiments on measured datasets validate improvement object using model.

Language: Английский

---

SOCR‐YOLO: Small Objects Detection Algorithm in Medical Images

International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(4)

Published: June 21, 2024

ABSTRACT In the field of medical image analysis, object detection plays a crucial role by providing interpretable diagnostic information to healthcare professionals. Although current models have achieved remarkable success in conventional images, their performance detecting abnormalities images has not been as satisfactory. This is primarily due complexity anatomical structures and fact that some lesions may subtle features, particularly case early‐stage, small‐scale abnormalities. To address this challenge, we introduce SOCR‐YOLO, novel lesion model with online convolutional reparameterization based on channel shuffling. First, it employs SOCR (Shuffled Channel Online Convolutional Re‐parameterization) module establish connection between feature concatenation computational efficiency, aiming extract more comprehensive while reducing time consumption. Second, incorporates Bi‐FPN structure achieve multiscale fusion. Lastly, loss function optimized improve training process. We evaluated two datasets, chest x‐ray (Vindr‐CXR) brain tumor (Br35H), provided Kaggle competition. Experimental results show proposed method outperformed several state‐of‐the‐art models, including YOLOv8, YOLO‐NAS, RT‐DETR, both speed accuracy. Notably, context anomaly detection, SOCR‐YOLO exhibits 1.8% enhancement accuracy over YOLOv8 simultaneously floating‐point operations 26.3%. Additionally, similar improvement observed tumors. The indicate superior ability our detect variations small lesions.

Language: Английский

---

Advancing common bean (Phaseolus vulgaris L.) disease detection with YOLO driven deep learning to enhance agricultural AI

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: July 6, 2024

Abstract Common beans (CB), a vital source for high protein content, plays crucial role in ensuring both nutrition and economic stability diverse communities, particularly Africa Latin America. However, CB cultivation poses significant threat to diseases that can drastically reduce yield quality. Detecting these solely based on visual symptoms is challenging, due the variability across different pathogens similar caused by distinct pathogens, further complicating detection process. Traditional methods relying farmers’ ability detect inadequate, while engaging expert pathologists advanced laboratories necessary, it also be resource intensive. To address this challenge, we present AI-driven system rapid cost-effective disease detection, leveraging state-of-the-art deep learning object technologies. We utilized an extensive image dataset collected from hotspots Colombia, focusing five major diseases: Angular Leaf Spot (ALS), Bacterial Blight (CBB), Bean Mosaic Virus (CBMV), Rust, Anthracnose, covering leaf pod samples real-field settings. images are only available disease. The study employed data augmentation techniques annotation at whole micro levels comprehensive analysis. train model, three YOLO architectures: YOLOv7, YOLOv8, YOLO-NAS. Particularly annotations, YOLO-NAS model achieves highest mAP value of up 97.9% recall 98.8%, indicating superior accuracy. In contrast, YOLOv7 YOLOv8 outperformed YOLO-NAS, with values exceeding 95% 93% recall. consistently yields lower performance than all classes plant parts, as examined models, highlighting unexpected discrepancy Furthermore, successfully deployed models into Android app, validating their effectiveness unseen classification accuracy (90%). This accomplishment showcases integration our production pipeline, process known DLOps. innovative approach significantly reduces diagnosis time, enabling farmers take prompt management interventions. potential benefits extend beyond serving early warning enhance common bean productivity

Language: Английский

---

Tiny-Object Detection Based on Optimized YOLO-CSQ for Accurate Drone Detection in Wildfire Scenarios

Drones, Journal Year: 2024, Volume and Issue: 8(9), P. 454 - 454

Published: Sept. 2, 2024

Wildfires, which are distinguished by their destructive nature and challenging suppression, present a significant threat to ecological environments socioeconomic systems. In order address this issue, the development of efficient accurate fire detection technologies for early warning timely response is essential. This paper addresses complexity forest mountain proposing YOLO-CSQ, drone-based method built upon an improved YOLOv8 algorithm. Firstly, we introduce CBAM attention mechanism, enhances model’s multi-scale feature extraction capabilities adaptively adjusting weights in both channel spatial dimensions maps, thereby improving accuracy. Secondly, propose ShuffleNetV2 backbone network structure, significantly reduces parameter count computational while maintaining capabilities. results more lightweight model. Thirdly, challenges varying scales numerous weak emission targets fires, Quadrupled-ASFF head weighted fusion. robustness detecting different scales. Finally, WIoU loss function replace traditional CIoU object function, enhancing localization The experimental demonstrate that model achieves mAP@50 96.87%, superior original YOLOV8, YOLOV9, YOLOV10 10.9, 11.66, 13.33 percentage points, respectively. Moreover, it exhibits advantages over other classic algorithms key evaluation metrics such as precision, recall, F1 score. These findings validate effectiveness scenarios, offering novel solution intelligent monitoring wildfires.

Language: Английский

---