Plant Detection and Counting: Enhancing Precision Agriculture in UAV and General Scenes

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 116196 - 116205

Published: Jan. 1, 2023

Plant detection and counting play a crucial role in modern agriculture, providing vital references for precision management resource allocation. This study follows the footsteps of machine learning experts by introducing state-of-the-art Yolov8 technology into field plant science. Moreover, we made some simple yet effective improvements. The integration shallow-level information path aggregation network (PANet) served to counterbalance resolution loss stemming from expanded receptive field. enhancement upsampled features was accomplished through combining lightweight up-sampling operator Content-Aware ReAssembly Features (CARAFE) with Multi-Efficient Channel Attention (Mlt-ECA) technique optimize features. collective approach markedly amplified discernment small objects Unmanned Aerial Vehicle (UAV) images, naming it Yolov8-UAV. Our evaluation is based on datasets containing four different species. Experimental results demonstrate strong competitiveness our proposed method even when compared most advanced techniques, possesses sufficient robustness. In order advance cross-disciplinary research computer vision science, also release new cotton boll dataset detailed annotated bounding box information. What's more, address previous oversights existing wheat ear updated labels consistent global advancements. Overall, this offers practitioners powerful solution addressing real-world application challenges. For UAV scenarios, recommend using specialized Yolov8-UAV, while Yolov8-N wise choice general scenes due its accuracy speed majority cases. Furthermore, contribute two meaningful that have significance, effectively promoting data resources short, contribution improve use scenarios open boxes. curated code can be accessed at following link: https://github.com/Ye-Sk/Plant-dataset.

Language: Английский

---

Accurate and fast implementation of soybean pod counting and localization from high-resolution image

Frontiers in Plant Science, Journal Year: 2024, Volume and Issue: 15

Published: Feb. 20, 2024

Introduction Soybean pod count is one of the crucial indicators soybean yield. Nevertheless, due to challenges associated with counting pods, such as crowded and uneven distribution, existing models prioritize accuracy over efficiency, which does not meet requirements for lightweight real-time tasks. Methods To address this goal, we have designed a deep convolutional network called PodNet. It employs encoder an efficient decoder that effectively decodes both shallow information, alleviating indirect interactions caused by information loss degradation between non-adjacent levels. Results We utilized high-resolution dataset pods from field harvesting evaluate model’s generalization ability. Through experimental comparisons manual model yield estimation, confirmed effectiveness PodNet model. The results indicate achieves R 2 0.95 prediction quantities compared ground truth, only 2.48M parameters, order magnitude lower than current SOTA YOLO POD, FPS much higher POD. Discussion Compared advanced computer vision methods, significantly enhances efficiency almost no sacrifice in accuracy. Its architecture high make it suitable applications, providing new solution locating dense objects.

Language: Английский

---

Fusing Global and Local Information Network for Tassel Detection in UAV Imagery

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 4100 - 4108

Published: Jan. 1, 2024

Unmanned aerial vehicles (UAVs), equipped with sensors, have made a significant impact in the field of agricultural analysis. Maize, being one most vital crops worldwide, is intricately linked to its yield and growth tassels. Leveraging UAV imagery for automatic monitoring maize tassels holds potential drive development intelligent cultivation. Current research methods, nevertheless, are limited lack robustness. To address challenge tassel detection images, we propose an innovative network, termed FGLNet. This network models backbone 16x down-sampling retain richer pixel information enhances performance by effectively fusing global local through weighted mechanisms. Moreover, scarcity data presents substantial constraint. In this study, publicly release new dataset, named counting (MTDC-UAV), featuring annotated bounding boxes, advance domain. Although images pose formidable challenges, our approach demonstrates remarkable accuracy evaluations based on MTDC-UAV dataset. It achieves AP ₅₀ 0.837 R ² 0.9409, all while maintaining parameter count just 0.77M. level considerably outperforms other state-of-the-art computer vision methods. Overall, not only introduces concepts but also provides worthwhile references solid foundation future studies.

Language: Английский

---

Improved YOLO-FastestV2 wheat spike detection model based on a multi-stage attention mechanism with a LightFPN detection head

Frontiers in Plant Science, Journal Year: 2024, Volume and Issue: 15

Published: June 19, 2024

The number of wheat spikes has an important influence on yield, and the rapid accurate detection spike numbers is great significance for yield estimation food security. Computer vision machine learning have been widely studied as potential alternatives to human detection. However, models with high accuracy are computationally intensive time consuming, lightweight tend lower precision. To address these concerns, YOLO-FastestV2 was selected base model comprehensive study analysis sheaf In this study, we constructed a target dataset comprising 11,451 images 496,974 bounding boxes. based Global Wheat Detection Dataset Sheaf Dataset, which published by PP Flying Paddle. We three attention mechanisms, Large Separable Kernel Attention (LSKA), Efficient Channel (ECA), Multi-Scale (EMA), enhance feature extraction capability backbone network improve underlying model. First, mechanism added after output phases network. Second, that further improved construct two-phase mechanism. On other hand, SimLightFPN introducing SimConv LightFPN module. results showed YOLO-FastestV2-SimLightFPN-ECA-EMA hybrid model, incorporates ECA in stage introduces EMA combination modules stage, best overall performance. P=83.91%, R=78.35%, AP= 81.52%, F1 = 81.03%, it ranked first GPI (0.84) evaluation. research examines deployment ear counting devices constrained resources, delivering novel solutions evolution agricultural automation precision agriculture.

Language: Английский

---

One to All: Towards a Unified Model for Counting Cereal Crop Heads Based on Few-Shot Learning

Plant Phenomics, Journal Year: 2024, Volume and Issue: 6

Published: Jan. 1, 2024

Accurate counting of cereals crops, e.g., maize, rice, sorghum, and wheat, is crucial for estimating grain production ensuring food security. However, existing methods cereal crops focus predominantly on building models specific crop head; thus, they lack generalizability to different varieties. This paper presents Counting Heads Cereal Crops Net (CHCNet), which a unified model designed multiple heads by few-shot learning, effectively reduces labeling costs. Specifically, refined vision encoder developed enhance feature embedding, where foundation model, namely, the segment anything (SAM), employed emphasize marked while mitigating complex background effects. Furthermore, multiscale interaction module proposed integrating similarity metric facilitate automatic learning crop-specific features across varying scales, enhances ability describe various sizes shapes. The CHCNet adopts 2-stage training procedure. initial stage focuses latent mining capture common representations crops. In subsequent stage, inference performed without additional training, extracting domain-specific target from selected exemplars accomplish task. extensive experiments 6 diverse datasets captured ground cameras drones, substantially outperformed state-of-the-art in terms cross-crop generalization ability, achieving mean absolute errors (MAEs) 9.96 9.38 13.94 7.94 15.62 mixed A user-friendly interactive demo available at http://cerealcropnet.com/, researchers are invited personally evaluate CHCNet. source code implementing https://github.com/Small-flyguy/CHCNet.

Language: Английский

---

WH-DETR: An Efficient Network Architecture for Wheat Spike Detection in Complex Backgrounds

Agriculture, Journal Year: 2024, Volume and Issue: 14(6), P. 961 - 961

Published: June 19, 2024

Wheat spike detection is crucial for estimating wheat yields and has a significant impact on the modernization of cultivation advancement precision agriculture. This study explores application DETR (Detection Transformer) architecture in detection, introducing new perspective to this task. We propose high-precision end-to-end network named WH-DETR, which based an enhanced RT-DETR architecture. Initially, we employ data augmentation techniques such as image rotation, scaling, random occlusion GWHD2021 dataset improve model’s generalization across various scenarios. A lightweight feature pyramid, GS-BiFPN, implemented network’s neck section effectively extract multi-scale features spikes complex environments, those with occlusions, overlaps, extreme lighting conditions. Additionally, introduction GSConv enhances while reducing computational costs, thereby controlling speed. Furthermore, EIoU metric integrated into loss function, refined better focus partially occluded or overlapping spikes. The testing results demonstrate that method achieves Average Precision (AP) 95.7%, surpassing current state-of-the-art object methods both These findings confirm our approach more closely meets practical requirements compared existing methods.

Language: Английский

---

SLFCNet: an ultra-lightweight and efficient strawberry feature classification network

PeerJ Computer Science, Journal Year: 2025, Volume and Issue: 11, P. e2085 - e2085

Published: Jan. 2, 2025

As modern agricultural technology advances, the automated detection, classification, and harvesting of strawberries have become an inevitable trend. Among these tasks, classification stands as a pivotal juncture. Nevertheless, existing object detection methods struggle with substantial computational demands, high resource utilization, reduced efficiency. These challenges make deployment on edge devices difficult lead to suboptimal user experiences. In this study, we developed lightweight model capable real-time strawberry fruit, named Strawberry Lightweight Feature Classify Network (SLFCNet). This innovative system incorporates encoder self-designed feature extraction module called Combined Convolutional Concatenation Sequential (C3SC). While maintaining compactness, architecture significantly enhances its decoding capabilities. To evaluate model's generalization potential, utilized high-resolution dataset collected directly from fields. By employing image augmentation techniques, conducted experimental comparisons between manually counted data inference-based results. The SLFCNet achieves average precision 98.9% in [email protected] metric, rate 94.7% recall 93.2%. Notably, features streamlined design, resulting compact size only 3.57 MB. On economical GTX 1080 Ti GPU, processing time per is mere 4.1 ms. indicates that can smoothly run devices, ensuring performance. Thus, it emerges novel solution for automation management harvesting, providing performance presenting new automatic picking.

Language: Английский

---

A Transformer-Based Symmetric Diffusion Segmentation Network for Wheat Growth Monitoring and Yield Counting

Agriculture, Journal Year: 2025, Volume and Issue: 15(7), P. 670 - 670

Published: March 21, 2025

A wheat growth and counting analysis model based on instance segmentation is proposed in this study to address the challenges of monitoring yield prediction high-density agricultural environments. The integrates transformer architecture with a symmetric attention mechanism employs diffusion module for precise measurement instances. By introducing an aggregated loss function, effectively optimizes both accuracy performance. Experimental results show that excels across several evaluation metrics. Specifically, task, using achieved Precision 0.91, Recall 0.87, Accuracy 0.89, mAP@75 0.88, F1-score significantly outperforming other baseline methods. For model’s reached 0.95, was 0.90, 0.93, 0.92, demonstrating marked advantage monitoring. Finally, provides novel effective method environments, offering substantial support future intelligent decision-making systems.

Language: Английский

---

PlantBiCNet: A new paradigm in plant science with bi-directional cascade neural network for detection and counting

Engineering Applications of Artificial Intelligence, Journal Year: 2023, Volume and Issue: 130, P. 107704 - 107704

Published: Dec. 30, 2023

Language: Английский

---

Multi-Altitude Corn Tassel Detection and Counting Based on UAV RGB Imagery and Deep Learning

Drones, Journal Year: 2024, Volume and Issue: 8(5), P. 198 - 198

Published: May 14, 2024

In the context of rapidly advancing agricultural technology, precise and efficient methods for crop detection counting play a crucial role in enhancing productivity efficiency management. Monitoring corn tassels is key to assessing plant characteristics, tracking health, predicting yield, addressing issues such as pests, diseases, nutrient deficiencies promptly. This ultimately ensures robust high-yielding growth. study introduces method recognition tassels, using RGB imagery captured by unmanned aerial vehicles (UAVs) YOLOv8 model. The model incorporates Pconv local convolution module, enabling lightweight design rapid speed. ACmix module added backbone section improve feature extraction capabilities tassels. Moreover, CTAM integrated into neck enhance semantic information exchange between channels, allowing positioning To optimize learning rate strategy, sparrow search algorithm (SSA) utilized. Significant improvements accuracy, efficiency, robustness are observed across various UAV flight altitudes. Experimental results show that, compared original model, proposed exhibits an increase accuracy 3.27 percentage points 97.59% recall 2.85 94.40% at height 5 m. Furthermore, optimizes frames per second (FPS), parameters (params), GFLOPs (giga floating point operations second) 7.12%, 11.5%, 8.94%, respectively, achieving values 40.62 FPS, 14.62 MB, 11.21 GFLOPs. At heights 10, 15, 20 m, maintains stable accuracies 90.36%, 88.34%, 84.32%, respectively. offers technical support automated intelligence precision production significantly contributing development modern technology.

Language: Английский

---