TIA-YOLOv5: An improved YOLOv5 network for real-time detection of crop and weed in the field

Frontiers in Plant Science, Год журнала: 2022, Номер 13

Опубликована: Дек. 22, 2022

Development of weed and crop detection algorithms provides theoretical support for control becomes an effective tool the site-specific management. For object tasks in field, there is often a large difference between number crop, resulting unbalanced distribution samples further posing difficulties task. In addition, most developed models tend to miss small objects, leading unsatisfied results. To overcome these issues, we proposed pixel-level synthesization data augmentation method TIA-YOLOv5 network complex field environment.The generated synthetic images by pasting pixels into original images. TIA-YOLOv5, transformer encoder block was added backbone improve sensitivity model weeds, channel feature fusion with involution (CFFI) strategy while reducing information loss, adaptive spatial (ASFF) introduced different scales prediction head.Test results publicly available sugarbeet dataset showed that yielded F1-scoreweed, APweed [email protected] 70.0%, 80.8% 90.0%, respectively, which 11.8%, 11.3% 5.9% higher than baseline YOLOv5 model. And speed reached 20.8 FPS.In this paper, fast accurate workflow including real-time field. The improved accuracy speed, providing very promising

Язык: Английский

---

SE-YOLOv5x: An Optimized Model Based on Transfer Learning and Visual Attention Mechanism for Identifying and Localizing Weeds and Vegetables

Agronomy, Год журнала: 2022, Номер 12(9), С. 2061 - 2061

Опубликована: Авг. 29, 2022

Weeds in the field affect normal growth of lettuce crops by competing with them for resources such as water and sunlight. The increasing costs weed management limited herbicide choices are threatening profitability, yield, quality lettuce. application intelligent weeding robots is an alternative to control intra-row weeds. prerequisite automatic accurate differentiation rapid localization different plants. In this study, a squeeze-and-excitation (SE) network combined You Only Look Once v5 (SE-YOLOv5x) proposed weed-crop classification field. Compared models including classical support vector machines (SVM), YOLOv5x, single-shot multibox detector (SSD), faster-RCNN, SE-YOLOv5x exhibited highest performance plant identifications, precision, recall, mean average precision (mAP), F1-score values 97.6%, 95.6%, 97.1%, 97.3%, respectively. Based on morphological characteristics, model detected location stem emerging points accuracy 97.14%. This study demonstrates capability weeds lettuce, which provides theoretical technical automated control.

Язык: Английский

---

Improved Apple Fruit Target Recognition Method Based on YOLOv7 Model

Agriculture, Год журнала: 2023, Номер 13(7), С. 1278 - 1278

Опубликована: Июнь 21, 2023

This study proposes an improved algorithm based on the You Only Look Once v7 (YOLOv7) to address low accuracy of apple fruit target recognition caused by high density, occlusion, and overlapping issues. Firstly, we proposed a preprocessing for split image with improve robotic intelligent picking accuracy. Then, divided training, validation, test sets. Secondly, MobileOne module was introduced into backbone network YOLOv7 achieve parametric fusion reduce computation. Afterward, SPPCSPS changed serial channel parallel enhance speed feature fusion. We added auxiliary detection head structure. Finally, conducted model validation tests. The results showed that increased 6.9%. recall rate 10%, mAP1 5%, mAP2 3.8%. 3.5%, 14%, 9.1%, 6.5% higher than other control YOLO algorithms, verifying could significantly in high-density fruits.

Язык: Английский

---

A comprehensive review of machine vision systems and artificial intelligence algorithms for the detection and harvesting of agricultural produce

Scientific African, Год журнала: 2023, Номер 21, С. e01798 - e01798

Опубликована: Июль 7, 2023

Every nation's economic development depends heavily on agriculture. Fulfilling the current population's need for food is becoming increasingly difficult because of factors including population growth, frequent climate change, and a lack resources. However, agriculture sector's biggest problems are trained workers, urbanization, available labour. Automation in essential to provide food, fibre, fuels rapidly growing population. Since harvesting critical step farming, authors present systematic review machine vision systems artificial intelligence algorithms detecting agricultural produce this article. The areas that being concentrated include systems, sensors, different image processing utilized detection harvesting. Review various types sensors used automated It demonstrates how several 3D methods, which were obtain position, orientation, point cloud fruit or crop, function compare them. Furthermore, it compares deployed precision This article shows knowledge-based can boost quality.

Язык: Английский

---

Agricultural weed identification in images and videos by integrating optimized deep learning architecture on an edge computing technology

Computers and Electronics in Agriculture, Год журнала: 2023, Номер 216, С. 108442 - 108442

Опубликована: Дек. 9, 2023

Язык: Английский

---

WeedVision: A single-stage deep learning architecture to perform weed detection and segmentation using drone-acquired images

Computers and Electronics in Agriculture, Год журнала: 2024, Номер 219, С. 108792 - 108792

Опубликована: Март 2, 2024

Deep learning (DL) inspired models have achieved tremendous success in locating target weed species through bounding-box approach (single-stage models) or pixel-wise semantic segmentation (two-stage models), but not both. Therefore, the goal of this research study was to develop a single-stage DL architecture that only locate presence bounding-boxes also achieves instance on unmanned aerial system (UAS) acquired remote sensing images. Moreover, developed experiments integrating novel C3 and C3x module within its backbone for dense feature extraction, as well ProtoNet (Prototypical network) head component masking. Furthermore, proposed has been trained five categories dataset exported using multiple combinations various augmentation techniques, namely, C1, C2, C3, C4, C5, which metrics were assessed desktop graphical processing unit (GPU) palm-sized edge device (AGX Xavier). Results suggest category combination six data outperformed remaining by achieving precision scores 85.4 % (bounding-boxes) 82.8 (masking) GPU. Whereas, same model converted TorchScript able achieve 79.1 77 masking accuracy an device, respectively. The two potential applications when integrated with site-specific management technologies. First, it enables real-time detection, allowing immediate identification weeds spot-spraying applications. Second, facilitates masking, aiding estimation growth extent actual field conditions. combines both computer vision - detection – provide comprehensive information about growth, eliminating need algorithm.

Язык: Английский

---

Real-time lettuce-weed localization and weed severity classification based on lightweight YOLO convolutional neural networks for intelligent intra-row weed control

Computers and Electronics in Agriculture, Год журнала: 2024, Номер 226, С. 109404 - 109404

Опубликована: Сен. 4, 2024

Язык: Английский

---

Railway obstacle intrusion warning mechanism integrating YOLO-based detection and risk assessment

Journal of Industrial Information Integration, Год журнала: 2024, Номер 38, С. 100571 - 100571

Опубликована: Янв. 11, 2024

Язык: Английский

---

Exploring the potential of visual tracking and counting for trees infected with pine wilt disease based on improved YOLOv5 and StrongSORT algorithm

Computers and Electronics in Agriculture, Год журнала: 2024, Номер 218, С. 108671 - 108671

Опубликована: Фев. 7, 2024

Язык: Английский

---

Assessing the Current and Future Potential Distribution of Solanum rostratum Dunal in China Using Multisource Remote Sensing Data and Principal Component Analysis

Remote Sensing, Год журнала: 2024, Номер 16(2), С. 271 - 271

Опубликована: Янв. 10, 2024

Accurate information concerning the spatial distribution of invasive alien species’ habitats is essential for species prevention and management, ecological sustainability. Currently, nationwide identification suitable highly destructive potentially weed, Solanum rostratum Dunal (S. rostratum), poses a series challenges. Simultaneously, research on potential future invasion areas likely directions spread has not received adequate attention. This study, based occurrence data multi-dimensional environmental variables constructed from multi-source remote sensing data, utilized Principal Component Analysis (PCA) in combination with Maxent model to effectively current habitat S. China, while quantitatively assessing various factors influencing its distribution. Research findings indicate that area covers 1.3952 million km2, all which located northern China. As trend climate warming persists, suitability range projected shift southward expand future; still predominantly it will have varying degrees expansion at different time frames. Notably, during period 2040 2061, under SSP1-2.6 scenario, exhibits most significant increase, surpassing scenario by 19.23%. Furthermore, attribution analysis PCA inverse transformation reveals soil, climate, spatial, humanistic, topographic collectively influence habitats, soil factors, particular, playing dominant role contributing up 75.85%. study identifies target management control rostratum, providing valuable insights into factor selection variable screening methods modeling (SDM).

Язык: Английский

---