A Lightweight Algorithm for Detecting Smoke in Forests without Open Flames DOI
Haowen Wang, Yan Piao, Yue Wang

et al.

2022 IEEE 4th International Conference on Civil Aviation Safety and Information Technology (ICCASIT), Journal Year: 2023, Volume and Issue: unknown, P. 201 - 205

Published: Oct. 11, 2023

Fast and accurate judgment of forest fire is great significance to prevention. Most the existing smoke detection models are only applicable case where there an open in image, excessive model volume makes it difficult be applied edge devices. To address this problem, a lightweight algorithm without proposed. The introduces attention mechanism CA full convolutional mask self-encoder framework FCMAE backbone network, so that can efficiently extract semantic information high low level features while solving feature collapse problem models. A centralized pyramid CFP also introduced prediction network enhance intra-layer conditioning features. In addition, uses loss function Wise-IoU with dynamic non-monotonic FM strengthen ability low-quality samples. Experimental results show has best performance detecting flame compared other

Language: Английский

An improved lightweight object detection algorithm for YOLOv5 DOI Creative Commons
Hao Luo, Jiangshu Wei, Yuchao Wang

et al.

PeerJ Computer Science, Journal Year: 2024, Volume and Issue: 10, P. e1830 - e1830

Published: Jan. 30, 2024

Object detection based on deep learning has made great progress in the past decade and been widely used various fields of daily life. Model lightweighting is core deploying target models mobile or edge devices. Lightweight have fewer parameters lower computational costs, but are often accompanied by accuracy. Based YOLOv5s, this article proposes an improved lightweight model, which can achieve higher accuracy with smaller parameters. Firstly, utilizing feature Ghost module, we integrated it into C3 structure replaced some modules after upsample layer neck network, thereby reducing number model expediting model’s inference process. Secondly, coordinate attention (CA) mechanism was added to enhance ability pay relevant information Finally, a more efficient Simplified Spatial Pyramid Pooling—Fast (SimSPPF) module designed stability shorten training time model. In order verify effectiveness experiments were conducted using three datasets different features. Experimental results show that our significantly reduced 28% compared original mean average precision (mAP) increased 3.1%, 1.1% 1.8% respectively. The also performs better terms existing state-of-the-art models. On features, mAP proposed achieved 87.2%, 77.8% 92.3%, than YOLOv7tiny (81.4%, 77.7%, 90.3%), YOLOv8n (84.7%, 90.6%) other advanced When achieving decreased parameters, successfully increase mAP, providing reference for

Language: Английский

Citations

2
Detection of variety and wax bloom of Shaanxi plum during post-harvest handling DOI Creative Commons
Hanchi Liu, Jinrong He,

Xuanping Fan

et al.

Chemometrics and Intelligent Laboratory Systems, Journal Year: 2024, Volume and Issue: 246, P. 105066 - 105066

Published: Jan. 20, 2024

The detection of plum variety and wax bloom has extensive applications in the fields fruit classification quality assessment. By automating identification varieties bloom, it is possible to enhance efficiency accuracy assessment, reduce manual intervention misjudgment, thereby improving market competitiveness fruits. Currently, many works focus on performance single attribute or often necessary use two models detect same information separately, which leads inefficient resource-consuming problems practical applications. To solve this problem improve detection, a Multi-Label model based YOLOv7 proposed. Firstly, double head structure introduced prediction ability for types features. Then, loss function suitable multi-attribute labels improved, functions are used optimize results labels, respectively. Finally, multi-label non-maximum suppression algorithm proposed filtering redundant bounding boxes labels. Experimental image dataset show that achieves [email protected] 96.2 %, precision 94.6 recall 89.5 %. experimental can effectively attributes label detection. code experiment be found at https://github.com/hejinrong/Muti-Label-YOLOv7.

Language: Английский

Citations

1
Implementation and Evaluation of Spatial Attention Mechanism in Apricot Disease Detection Using Adaptive Sampling Latent Variable Network DOI Creative Commons

Bingyuan Han,

Peiyan Duan,

Chengcheng Zhou

et al.

Plants, Journal Year: 2024, Volume and Issue: 13(12), P. 1681 - 1681

Published: June 18, 2024

In this study, an advanced method for apricot tree disease detection is proposed that integrates deep learning technologies with various data augmentation strategies to significantly enhance the accuracy and efficiency of detection. A comprehensive framework based on adaptive sampling latent variable network (ASLVN) spatial state attention mechanism was developed aim enhancing model’s capability capture characteristics diseases while ensuring its applicability edge devices through model lightweighting techniques. Experimental results demonstrated significant improvements in precision, recall, accuracy, mean average precision (mAP). Specifically, 0.92, recall 0.89, 0.90, mAP 0.91, surpassing traditional models such as YOLOv5, YOLOv8, RetinaNet, EfficientDet, DEtection TRansformer (DETR). Furthermore, ablation studies, critical roles ASLVN performance were validated. These experiments not only showcased contributions each component improving but also highlighted method’s address challenges complex environments. Eight types detected, including Powdery Mildew Brown Rot, representing a technological breakthrough. The findings provide robust technical support management actual agricultural production offer broad application prospects.

Language: Английский

Citations

1
YOLO-Chili: An Efficient Lightweight Network Model for Localization of Pepper Picking in Complex Environments DOI Creative Commons
Hailin Chen, Ruofan Zhang, Jialiang Peng

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(13), P. 5524 - 5524

Published: June 25, 2024

Currently, few deep models are applied to pepper-picking detection, and existing generalized neural networks face issues such as large model parameters, prolonged training times, low accuracy. To address these challenges, this paper proposes the YOLO-chili target detection algorithm for chili pepper detection. Initially, classical YOLOv5 serves benchmark model. We introduce an adaptive spatial feature pyramid structure that combines attention mechanism concept of multi-scale prediction enhance model’s capabilities occluded small peppers. Subsequently, we incorporate a three-channel module improve algorithm’s long-distance recognition ability reduce interference from redundant objects. Finally, employ quantized pruning method parameters achieve lightweight processing. Applying our custom dataset, average precision (AP) value 93.11% with accuracy rate 93.51% recall 92.55%. The experimental results demonstrate enables accurate real-time in complex orchard environments.

Language: Английский

Citations

1
AgRegNet: A deep regression network for flower and fruit density estimation, localization, and counting in orchards DOI
Uddhav Bhattarai,

Santosh Bhusal,

Qin Zhang

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 227, P. 109534 - 109534

Published: Oct. 24, 2024

Language: Английский

Citations

1
Multi-scale convolution and dynamic task interaction detection head for efficient lightweight plum detection DOI

Jiachun Wu,

Jinlai Zhang, Jihong Zhu

et al.

Food and Bioproducts Processing, Journal Year: 2024, Volume and Issue: 149, P. 353 - 367

Published: Dec. 11, 2024

Language: Английский

Citations

0
Research on detection of wheat tillers in natural environment based on YOLOv8-MRF DOI Creative Commons
Min Liang, Yuchen Zhang,

Jian Zhou

et al.

Smart Agricultural Technology, Journal Year: 2024, Volume and Issue: unknown, P. 100720 - 100720

Published: Dec. 1, 2024

Language: Английский

Citations

0
A Lightweight Algorithm for Detecting Smoke in Forests without Open Flames DOI
Haowen Wang, Yan Piao, Yue Wang

et al.

2022 IEEE 4th International Conference on Civil Aviation Safety and Information Technology (ICCASIT), Journal Year: 2023, Volume and Issue: unknown, P. 201 - 205

Published: Oct. 11, 2023

Fast and accurate judgment of forest fire is great significance to prevention. Most the existing smoke detection models are only applicable case where there an open in image, excessive model volume makes it difficult be applied edge devices. To address this problem, a lightweight algorithm without proposed. The introduces attention mechanism CA full convolutional mask self-encoder framework FCMAE backbone network, so that can efficiently extract semantic information high low level features while solving feature collapse problem models. A centralized pyramid CFP also introduced prediction network enhance intra-layer conditioning features. In addition, uses loss function Wise-IoU with dynamic non-monotonic FM strengthen ability low-quality samples. Experimental results show has best performance detecting flame compared other

Language: Английский

Citations

0