MTS-YOLO: A Multi-Task Lightweight and Efficient Model for Tomato Fruit Bunch Maturity and Stem Detection DOI Creative Commons
Maonian Wu,

H.L. Lin,

Xingren Shi

et al.

Horticulturae, Journal Year: 2024, Volume and Issue: 10(9), P. 1006 - 1006

Published: Sept. 22, 2024

The accurate identification of tomato maturity and picking positions is essential for efficient picking. Current deep-learning models face challenges such as large parameter sizes, single-task limitations, insufficient precision. This study proposes MTS-YOLO, a lightweight model detecting fruit bunch stem positions. We reconstruct the YOLOv8 neck network propose high- low-level interactive screening path aggregation (HLIS-PAN), which achieves excellent multi-scale feature extraction through alternating fusion information while reducing number parameters. Furthermore, utilize DySample upsampling, bypassing complex kernel computations with point sampling. Moreover, context anchor attention (CAA) introduced to enhance model’s ability recognize elongated targets bunches stems. Experimental results indicate that MTS-YOLO an F1-score 88.7% [email protected] 92.0%. Compared mainstream models, not only enhances accuracy but also optimizes size, effectively computational costs inference time. precisely identifies foreground need be harvested ignoring background objects, contributing improved efficiency. provides technical solution intelligent agricultural

Language: Английский

EDGS-YOLOv8: An Improved YOLOv8 Lightweight UAV Detection Model DOI Creative Commons
Min Huang,

Wenkai Mi,

Yuming Wang

et al.

Drones, Journal Year: 2024, Volume and Issue: 8(7), P. 337 - 337

Published: July 20, 2024

In the rapidly developing drone industry, use has led to a series of safety hazards in both civil and military settings, making detection an increasingly important research field. It is difficult overcome this challenge with traditional object solutions. Based on YOLOv8, we present lightweight, real-time, accurate anti-drone model (EDGS-YOLOv8). This performed by improving structure, introducing ghost convolution neck reduce size, adding efficient multi-scale attention (EMA), head using DCNv2 (deformable convolutional net v2). The proposed method evaluated two UAV image datasets, DUT Anti-UAV Det-Fly, comparison YOLOv8 baseline model. results demonstrate that dataset, EDGS-YOLOv8 achieves AP value 0.971, which 3.1% higher than YOLOv8n’s mAP, while maintaining size only 4.23 MB. findings methods outlined here are crucial for target accuracy lightweight models.

Language: Английский

Citations

11
Improved YOLOv8 Model for Lightweight Pigeon Egg Detection DOI Creative Commons
Tao Jiang, Jie Zhou, Binbin Xie

et al.

Animals, Journal Year: 2024, Volume and Issue: 14(8), P. 1226 - 1226

Published: April 19, 2024

In response to the high breakage rate of pigeon eggs and significant labor costs associated with egg-producing farming, this study proposes an improved YOLOv8-PG (real versus fake egg detection) model based on YOLOv8n. Specifically, Bottleneck in C2f module YOLOv8n backbone network neck are replaced Fasternet-EMA Block Fasternet Block, respectively. The is designed PConv (Partial Convolution) reduce parameter count computational load efficiently. Furthermore, incorporation EMA (Efficient Multi-scale Attention) mechanism helps mitigate interference from complex environments pigeon-egg feature-extraction capabilities. Additionally, Dysample, ultra-lightweight effective upsampler, introduced into further enhance performance lower overhead. Finally, EXPMA (exponential moving average) concept employed optimize SlideLoss propose EMASlideLoss classification loss function, addressing issue imbalanced data samples enhancing model's robustness. experimental results showed that F1-score, mAP50-95, mAP75 increased by 0.76%, 1.56%, 4.45%, respectively, compared baseline model. Moreover, reduced 24.69% 22.89%, Compared detection models such as Faster R-CNN, YOLOv5s, YOLOv7, YOLOv8s, exhibits superior performance. reduction contributes lowering deployment facilitates its implementation mobile robotic platforms.

Language: Английский

Citations

9
Picking point localization method of table grape picking robot based on you only look once version 8 nano DOI
Yanjun Zhu,

Shunshun Sui,

Wensheng Du

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 146, P. 110266 - 110266

Published: Feb. 15, 2025

Language: Английский

Citations

1
GPC-YOLO: An Improved Lightweight YOLOv8n Network for the Detection of Tomato Maturity in Unstructured Natural Environments DOI Creative Commons
Yafeng Dong, Jinwei Qiao, Na Liu

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(5), P. 1502 - 1502

Published: Feb. 28, 2025

Effective fruit identification and maturity detection are important for harvesting managing tomatoes. Current deep learning algorithms typically demand significant computational resources memory. Detecting severely stacked obscured tomatoes in unstructured natural environments is challenging because of target stacking, occlusion, illumination, background noise. The proposed method involves a new lightweight model called GPC-YOLO based on YOLOv8n tomato detection. This study proposes C2f-PC module partial convolution (PConv) less computation, which replaced the original C2f feature extraction YOLOv8n. regular was with Grouped Spatial Convolution (GSConv) by downsampling to reduce burden. neck network convolutional neural network-based cross-scale fusion (CCFF) enhance adaptability scale changes detect many small-scaled objects. Additionally, integration simple attention mechanism (SimAM) efficient intersection over union (EIoU) loss were implemented further accuracy leveraging these improvements. trained validated dataset 1249 mobile phone images Compared YOLOv8n, achieved high-performance metrics, e.g., reducing parameter number 1.2 M (by 59.9%), compressing size 2.7 57.1%), decreasing floating point operations 4.5 G 45.1%), improving 98.7% 0.3%), speed 201 FPS. showed that could effectively identify environments. has immense potential ripeness automated picking applications.

Language: Английский

Citations

1
YOLOv10-pose and YOLOv9-pose: Real-time strawberry stalk pose detection models DOI

Zhichao Meng,

Xiaoqiang Du, Ranjan Sapkota

et al.

Computers in Industry, Journal Year: 2024, Volume and Issue: 165, P. 104231 - 104231

Published: Dec. 19, 2024

Language: Английский

Citations

6
Simultaneous detection of fruits and fruiting stems in mango using improved YOLOv8 model deployed by edge device DOI
Zenan Gu, Deqiang He, Junduan Huang

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 227, P. 109512 - 109512

Published: Oct. 2, 2024

Language: Английский

Citations

5
Cognition of grape cluster picking point based on visual knowledge distillation in complex vineyard environment DOI
Jinhai Wang, Xuemin Lin, Lufeng Luo

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 225, P. 109216 - 109216

Published: July 30, 2024

Language: Английский

Citations

4
Research on a Trellis Grape Stem Recognition Method Based on YOLOv8n-GP DOI Creative Commons
Tong Jiang, Yane Li, Hailin Feng

et al.

Agriculture, Journal Year: 2024, Volume and Issue: 14(9), P. 1449 - 1449

Published: Aug. 25, 2024

Grapes are an important cash crop that contributes to the rapid development of agricultural economy. The harvesting ripe fruits is one crucial steps in grape production process. However, at present, picking methods mainly manual, resulting wasted time and high costs. Therefore, it particularly implement intelligent picking, which accurate detection stems a key step achieve harvesting. In this study, trellis stem model, YOLOv8n-GP, was proposed by combining SENetV2 attention module CARAFE upsampling operator with YOLOv8n-pose. Specifically, study first embedded bottom backbone network enhance model’s ability extract feature information. Then, we utilized replace modules neck network, expanding sensory field model without increasing its parameters. Finally, validate performance examined effectiveness various keypoint models constructed YOLOv8n-pose, YOLOv5-pose, YOLOv7-pose, YOLOv7-Tiny-pose. Experimental results show precision, recall, mAP, mAP-kp YOLOv8n-GP reached 91.6%, 91.3%, 97.1%, 95.4%, improved 3.7%, 3.6%, 4.6%, 4.0%, respectively, compared Furthermore, exhibits superior other terms each evaluation indicator. experimental demonstrate can detect efficiently accurately, providing technical support for advancing

Language: Английский

Citations

4
Positioning of mango picking point using an improved YOLOv8 architecture with object detection and instance segmentation DOI
Hongwei Li, Jianzhi Huang, Zenan Gu

et al.

Biosystems Engineering, Journal Year: 2024, Volume and Issue: 247, P. 202 - 220

Published: Sept. 23, 2024

Language: Английский

Citations

4
OW-YOLO: An Improved YOLOv8s Lightweight Detection Method for Obstructed Walnuts DOI Creative Commons
Haoyu Wang, Lijun Yun,

Chenggui Yang

et al.

Agriculture, Journal Year: 2025, Volume and Issue: 15(2), P. 159 - 159

Published: Jan. 13, 2025

Walnut detection in mountainous and hilly regions often faces significant challenges due to obstructions, which adversely affect model performance. To address this issue, we collected a dataset comprising 2379 walnut images from these regions, with detailed annotations for both obstructed non-obstructed walnuts. Based on dataset, propose OW-YOLO, lightweight object specifically designed detecting small, The model’s backbone was restructured the integration of DWR-DRB (Dilated Weighted Residual-Dilated Residual Block) module. enhance efficiency multi-scale feature fusion, incorporated HSFPN (High-Level Screening Feature Pyramid Network) redesigned head by replacing original more efficient LADH while removing processing 32 × maps. These improvements effectively reduced complexity significantly enhanced accuracy Experiments were conducted using PyTorch framework an NVIDIA GeForce RTX 4060 Ti GPU. results demonstrate that OW-YOLO outperforms other models, achieving [email protected] (mean average precision) 83.6%, mAP@[0.5:0.95] 53.7%, F1 score 77.9%. Additionally, parameter count decreased 49.2%, weight file size 48.1%, computational load dropped 37.3%, mitigating impact obstruction accuracy. findings provide robust support future development agriculture lay solid foundation broader adoption intelligent agriculture.

Language: Английский

Citations

0