An optimized Yolov5s approach and its application to image detection problems DOI

Jiayu Song,

Zhiquan Zhu,

Yufei Qiang

et al.

2022 IEEE 5th International Conference on Information Systems and Computer Aided Education (ICISCAE), Journal Year: 2023, Volume and Issue: unknown, P. 1113 - 1118

Published: Sept. 23, 2023

Traditional object detection algorithms cannot cope with motion blur and have poor classification accuracy for small cluster targets, which poses challenges to the development of image technology. In order address these issues, this paper adopts a multi-strategy integrated optimization framework optimizes training speed model based on YOLOv5s network model. Firstly, CBAM attention mechanism is introduced mitigate blur. Then, Focal-EIoU loss function used enhance low-accuracy objects. Next, SiLU activation employed model's representation capability, SGD optimizer replaced Adam improve efficiency. Finally, improved underwent 500 rounds subset BDD100K dataset. The results show that model, after multiple strategy optimizations, increased from 46.5% 74.7%. optimized by strategies, can effectively solve problems such as blur, thereby improving stability

Language: Английский

Research on automatic pavement crack identification Based on improved YOLOv8 DOI

Hongyu Wang,

Xiao Han,

Xifa Song

et al.

International Journal on Interactive Design and Manufacturing (IJIDeM), Journal Year: 2024, Volume and Issue: 18(6), P. 3773 - 3783

Published: Feb. 24, 2024

Language: Английский

Citations

11
Intelligent identification of rice leaf disease based on YOLO V5-EFFICIENT DOI
Weiwei Gao,

Chunxu Zong,

Manyi Wang

et al.

Crop Protection, Journal Year: 2024, Volume and Issue: 183, P. 106758 - 106758

Published: May 23, 2024

Language: Английский

Citations

11
Rice disease segmentation method based on CBAM-CARAFE-DeepLabv3+ DOI
Wei Zeng, Mingfang He

Crop Protection, Journal Year: 2024, Volume and Issue: 180, P. 106665 - 106665

Published: March 22, 2024

Language: Английский

Citations

10
Real-Time Detection of Varieties and Defects in Moving Corn Seeds Based on YOLO-SBWL DOI Creative Commons
Yuhang Che, Hongyi Bai, Laijun Sun

et al.

Agriculture, Journal Year: 2025, Volume and Issue: 15(7), P. 685 - 685

Published: March 24, 2025

Sorting corn seeds before sowing is crucial to ensure the varietal purity of and yield crop. However, most existing methods for sorting cannot detect both varieties defects simultaneously. Detecting in motion more difficult than at rest, many models pursue high accuracy expense model inference time. To address these issues, this study proposed a real-time detection model, YOLO-SBWL, that simultaneously identifies seed surface by using images taken different conveyor speeds. False damaged was addressed inserting simple parameter-free attention mechanism (SimAM) into original “you only look once” (YOLO)v7 network. At neck network, path-aggregation feature pyramid network replaced with weighted bi-directional (BiFPN) increase classifying undamaged seeds. The Wise-IoU loss function supplanted CIoU mitigate adverse impacts caused low-quality samples. Finally, improved pruned layer-adaptive magnitude-based pruning (LAMP) effectively compress model. YOLO-SBWL demonstrated mean average precision 97.21%, which 2.59% higher GFLOPs were reduced 67.16%, size decreased 67.21%. during belt movement remained above 96.17%, times within 11 ms. This provided technical support swift precise identification transport.

Language: Английский

Citations

1
Discriminating Spectral–Spatial Feature Extraction for Hyperspectral Image Classification: A Review DOI Creative Commons
Ningyang Li, Zhaohui Wang, Faouzi Alaya Cheikh

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(10), P. 2987 - 2987

Published: May 8, 2024

Hyperspectral images (HSIs) contain subtle spectral details and rich spatial contextures of land cover that benefit from developments in imaging space technology. The classification HSIs, which aims to allocate an optimal label for each pixel, has broad prospects the field remote sensing. However, due redundancy between bands complex structures, effectiveness shallow spectral–spatial features extracted by traditional machine-learning-based methods tends be unsatisfying. Over recent decades, various based on deep learning computer vision have been proposed allow discrimination representations classification. In this article, crucial factors discriminate are systematically summarized perspectives feature extraction optimization. For extraction, techniques ensure features, illustrated characteristics hyperspectral data architecture models. optimization, adjust distances classes introduced detail. Finally, limitations these future challenges facilitating HSI also discussed further.

Language: Английский

Citations

5
DBY-Tobacco: a dual-branch model for non-tobacco related materials detection based on hyperspectral feature fusion DOI Creative Commons
Cheng Shen,

Yuecheng Qi,

Lijun Yun

et al.

Frontiers in Plant Science, Journal Year: 2025, Volume and Issue: 16

Published: March 5, 2025

The removal of non-tobacco related materials (NTRMs) is crucial for improving tobacco product quality and consumer safety. Traditional NTRM detection methods are labor-intensive inefficient. This study proposes a novel approach real-time using hyperspectral imaging (HSI) an enhanced YOLOv8 model, named Dual-branch-YOLO-Tobacco (DBY-Tobacco). We created dataset 1,000 images containing 4,203 NTRMs by camera, SpectraEye (SEL-24), with spectral range 400-900 nm. To improve processing efficiency HSIs data, three characteristic wavelengths (580nm, 680nm, 850nm) were extracted analyzing the weighted coefficients principal components. Then pseudo color image fusion decorrelation contrast stretch applied enhancement. DBY-Tobacco model features dual-branch backbone network BiFPN-Efficient-Lighting-Feature-Pyramid-Network (BELFPN) module effective feature fusion. Experimental results demonstrate that achieves high performance metrics, including F1 score 89.7%, mAP@50 92.8%, mAP@50-95 73.7%, speed 151 FPS, making it suitable applications in dynamic production environments. highlights potential combining HSI advanced deep learning techniques Future work will focus on addressing limitations such as stripe noise expanding to other types NTRMs. code available at: https://github.com/Ikaros-sc/DBY-Tobacco.

Language: Английский

Citations

0
Deep learning-based target spraying control of weeds in wheat fields at tillering stage DOI Creative Commons
Haiying Wang, Yu Chen, Shuo Zhang

et al.

Frontiers in Plant Science, Journal Year: 2025, Volume and Issue: 16

Published: March 27, 2025

In this study, a target spraying decision and hysteresis algorithm is designed in conjunction with deep learning, which deployed on testbed for validation. The overall scheme of the control system first proposed. Then YOLOv5s lightweighted improved. Based this, designed, so that can precisely solenoid valve differentiate according to distribution weeds different areas, at same time, successfully solve operation problem between hardware. Finally, was simulated tillering wheat were selected bench experiments. Experiments dataset realistic scenarios show improved model reduces GFLOPs (computational complexity) size by 52.2% 42.4%, respectively, mAP F1 91.4% 85.3%, an improvement 0.2% 0.8%, compared original model. results experiments showed rate under speed intervals 0.3-0.4m/s, 0.4-0.5m/s 0.5-0.6m/s reached 99.8%, 98.2% 95.7%, respectively. Therefore, provide excellent accuracy performance system, thus laying theoretical foundation practical application spraying.

Language: Английский

Citations

0
Deciphering rice feralization: insights from genomics of weedy rice DOI

Yunqi Cong,

Yijie Gui,

Kaicheng Yong

et al.

Published: Jan. 1, 2025

Language: Английский

Citations

0
An efficient segmentation model for abnormal chicken droppings recognition based on improved deep dual-resolution network DOI
Pengguang He, Rui Wu, Da Liu

et al.

Journal of Animal Science, Journal Year: 2024, Volume and Issue: 102

Published: Jan. 1, 2024

The characteristics of chicken droppings are closely linked to their health status. In prior studies, recognition is treated as an object detection task, leading challenges in labeling and missed due the diverse shapes, overlapping boundaries, dense distribution droppings. Additionally, use intelligent monitoring equipment equipped with edge devices farms can significantly reduce manual labor. However, limited computational power presents deploying real-time segmentation algorithms for field applications. Therefore, this study redefines task a main objective being development lightweight model automated abnormal A total 60 Arbor Acres broilers were housed 5 specific pathogen-free cages over 3 wk, 1650 RGB images randomly divided into training testing sets 8:2 ratio develop test model. Firstly, by incorporating attention mechanism, multi-loss function, auxiliary head, accuracy DDRNet was enhanced. Then, employing group convolution advanced knowledge-distillation algorithm, named DDRNet-s-KD obtained, which achieved mean Dice coefficient (mDice) 79.43% inference speed 86.10 frames per second (FPS), showing 2.91% 61.2% increase mDice FPS compared benchmark Furthermore, quantized from 32-bit floating-point values 8-bit integers then converted TensorRT format. Impressively, weight size only 13.7 MB, representing 82.96% reduction This makes it well-suited deployment on device, achieving 137.51 Jetson Xavier NX. conclusion, methods proposed show significant potential provide effective reference implementation other agricultural embedded systems.

Language: Английский

Citations

3
Detecting rice (Oryza sativa) panicle using an improved YOLOv5 model DOI

Xiaoyue Seng,

Xue Yang, Tonghai Liu

et al.

Crop and Pasture Science, Journal Year: 2025, Volume and Issue: 76(2)

Published: Feb. 13, 2025

Context Rice (Oryza sativa) panicle provides important information to improve production efficiency, optimise resources, and aid in successful breeding of high-performing rice varieties. Aims In order efficiently count panicles, a recognition model based on YOLOv5s-Slim Neck-GhostNet was evaluated. Methods We used the developmental stages from heading maturity as time period collect data for testing validating model. The GSConv convolution module YOLOv5 (You Only Look Once) compared with original Conv convolution. improved C3 replaced it VoVGSCSP module, which further enhanced detection ability small targets, such panicles. To performance reduce computational complexity, we backbone network lightweight efficient GhostNet structure. Key results Our showed that precision test set 96.5%, recall 94.6%, F1-score 95.5%, [email protected] 97.2%. Compared YOLOv5s model, increased by 1.8%, size is reduced 5.7M. Conclusions had capability detect panicles real time. method while maintaining an acceptable level accuracy. Implications technology intelligent automated solution better monitor development, has potential practical application agricultural settings.

Language: Английский

Citations

0