A Lightweight Cotton Verticillium Wilt Hazard Level Real-Time Assessment System Based on an Improved YOLOv10n Model DOI Creative Commons
Juan Liao,

Xinying He,

Yexiong Liang

и другие.

Agriculture, Год журнала: 2024, Номер 14(9), С. 1617 - 1617

Опубликована: Сен. 14, 2024

Compared to traditional manual methods for assessing the cotton verticillium wilt (CVW) hazard level, utilizing deep learning models foliage segmentation can significantly improve evaluation accuracy. However, instance images with complex backgrounds often suffer from low accuracy and delayed segmentation. To address this issue, an improved model, YOLO-VW, high accuracy, efficiency, a light weight, was proposed CVW level assessment based on YOLOv10n model. (1) It replaced conventional convolutions lightweight GhostConv, reducing computational time. (2) The STC module Swin Transformer enhanced expression of disease spot boundary features, further model size. (3) integrated squeeze-and-excitation (SE) attention mechanism suppress irrelevant background information. (4) employed stochastic gradient descent (SGD) optimizer enhance performance shorten detection severity then deployed server, real-time application (APP) developed results indicated following. YOLO-VW achieved mean average precision (mAP) 89.2% frame per second (FPS) rate 157.98 f/s in CVW, representing improvements 2.4% 21.37 over original respectively. model’s parameters floating point operations (FLOPs) were 1.59 M 7.8 G, respectively, compressed by 44% 33.9% compared After deploying smartphone, processing time each image 2.42 s, under various environmental conditions reached 85.5%, 15% improvement Based these findings, meets requirements detection, offering greater robustness, portability practical applications. This provides technical support controlling developing varieties resistant wilt.

Язык: Английский

A Comprehensive Review of Deep Learning Applications in Cotton Industry: From Field Monitoring to Smart Processing DOI Creative Commons
Zheng Yang,

W. L. Xia,

Hone‐Jay Chu

и другие.

Plants, Год журнала: 2025, Номер 14(10), С. 1481 - 1481

Опубликована: Май 15, 2025

Cotton is a vital economic crop in global agriculture and the textile industry, contributing significantly to food security, industrial competitiveness, sustainable development. Traditional technologies such as spectral imaging machine learning improved cotton cultivation processing, yet their performance often falls short complex agricultural environments. Deep (DL), with its superior capabilities data analysis, pattern recognition, autonomous decision-making, offers transformative potential across value chain. This review highlights DL applications seed quality assessment, pest disease detection, intelligent irrigation, harvesting, fiber classification et al. enhances accuracy, efficiency, adaptability, promoting modernization of production precision agriculture. However, challenges remain, including limited model generalization, high computational demands, environmental adaptability issues, costly annotation. Future research should prioritize lightweight, robust models, standardized multi-source datasets, real-time optimization. Integrating multi-modal data-such remote sensing, weather, soil information-can further boost decision-making. Addressing these will enable play central role driving intelligent, automated, transformation industry.

Язык: Английский

Процитировано

0
A Lightweight Cotton Verticillium Wilt Hazard Level Real-Time Assessment System Based on an Improved YOLOv10n Model DOI Creative Commons
Juan Liao,

Xinying He,

Yexiong Liang

и другие.

Agriculture, Год журнала: 2024, Номер 14(9), С. 1617 - 1617

Опубликована: Сен. 14, 2024

Compared to traditional manual methods for assessing the cotton verticillium wilt (CVW) hazard level, utilizing deep learning models foliage segmentation can significantly improve evaluation accuracy. However, instance images with complex backgrounds often suffer from low accuracy and delayed segmentation. To address this issue, an improved model, YOLO-VW, high accuracy, efficiency, a light weight, was proposed CVW level assessment based on YOLOv10n model. (1) It replaced conventional convolutions lightweight GhostConv, reducing computational time. (2) The STC module Swin Transformer enhanced expression of disease spot boundary features, further model size. (3) integrated squeeze-and-excitation (SE) attention mechanism suppress irrelevant background information. (4) employed stochastic gradient descent (SGD) optimizer enhance performance shorten detection severity then deployed server, real-time application (APP) developed results indicated following. YOLO-VW achieved mean average precision (mAP) 89.2% frame per second (FPS) rate 157.98 f/s in CVW, representing improvements 2.4% 21.37 over original respectively. model’s parameters floating point operations (FLOPs) were 1.59 M 7.8 G, respectively, compressed by 44% 33.9% compared After deploying smartphone, processing time each image 2.42 s, under various environmental conditions reached 85.5%, 15% improvement Based these findings, meets requirements detection, offering greater robustness, portability practical applications. This provides technical support controlling developing varieties resistant wilt.

Язык: Английский

Процитировано

2