A Dragon Fruit Picking Detection Method Based on YOLOv7 and PSP-Ellipse

Sensors, Год журнала: 2023, Номер 23(8), С. 3803 - 3803

Опубликована: Апрель 7, 2023

Dragon fruit is one of the most popular fruits in China and Southeast Asia. It, however, mainly picked manually, imposing high labor intensity on farmers. The hard branches complex postures dragon make it difficult to achieve automated picking. For picking with diverse postures, this paper proposes a new detection method, not only identify locate fruit, but also detect endpoints that are at head root which can provide more visual information for robot. First, YOLOv7 used classify fruit. Then, we propose PSP-Ellipse method further including segmentation via PSPNet, positioning an ellipse fitting algorithm classification ResNet. To test proposed some experiments conducted. In detection, precision, recall average precision 0.844, 0.924 0.932, respectively. performs better compared other models. segmentation, performance PSPNet than commonly semantic models, mean intersection over union being 0.959, 0.943 0.906, distance error angle based 39.8 pixels 4.3°, accuracy ResNet 0.92. makes great improvement two kinds keypoint regression UNet. Orchard verified effective. promotes progress automatic provides reference detection.

Язык: Английский

---

RDE-YOLOv7: An Improved Model Based on YOLOv7 for Better Performance in Detecting Dragon Fruits

Agronomy, Год журнала: 2023, Номер 13(4), С. 1042 - 1042

Опубликована: Март 31, 2023

There is a great demand for dragon fruit in China and Southeast Asia. Manual picking of requires lot labor. It imperative to study the fruit-picking robot. The visual guidance system an important part To realize automatic fruit, this paper proposes detection method based on RDE-YOLOv7 identify locate more accurately. RepGhost decoupled head are introduced into YOLOv7 better extract features predict results. In addition, multiple ECA blocks various locations network effective information from large amount information. experimental results show that improves precision, recall, mean average precision by 5.0%, 2.1%, 1.6%. also has high accuracy under different lighting conditions blur degrees. Using RDE-YOLOv7, we build conduct positioning experiments. spatial error only 2.51 mm, 2.43 1.84 mm. experiments indicate can accurately detect fruits, theoretically supporting development robots.

Язык: Английский

---

An Automatic Jujube Fruit Detection and Ripeness Inspection Method in the Natural Environment

Agronomy, Год журнала: 2023, Номер 13(2), С. 451 - 451

Опубликована: Фев. 2, 2023

The ripeness phases of jujube fruits are one factor mitigating against fruit detection, in addition to uneven environmental conditions such as illumination variation, leaf occlusion, overlapping fruits, colors or brightness, similar plant appearance the background, and so on. Therefore, a method called YOLO-Jujube was proposed solve these problems. With incorporation networks Stem, RCC, Maxpool, CBS, SPPF, C3, PANet, CIoU loss, able detect automatically for inspection. Having recorded params 5.2 m, GFLOPs 11.7, AP 88.8%, speed 245 fps detection performance, including sorting counting process combined, outperformed network YOLOv3-tiny, YOLOv4-tiny, YOLOv5s, YOLOv7-tiny. is robust applicable meet goal computer vision-based understanding images videos.

Язык: Английский

---

YOLOv8-Peas: a lightweight drought tolerance method for peas based on seed germination vigor

Frontiers in Plant Science, Год журнала: 2023, Номер 14

Опубликована: Сен. 28, 2023

Drought stress has become an important factor affecting global food production. Screening and breeding new varieties of peas (Pisum sativum L.) for drought-tolerant is critical importance to ensure sustainable agricultural production security. Germination rate germination index are indicators seed vigor, the level vigor pea seeds directly affects their yield quality. The traditional manual detection can hardly meet demand full-time sequence nondestructive detection. We propose YOLOv8-Peas, improved YOLOv8-n based method vigor.We constructed a dataset used multiple data augmentation methods improve robustness model in real-world scenarios. By introducing C2f-Ghost structure depth-separable convolution, computational complexity reduced size compressed. In addition, original detector head replaced by self-designed PDetect head, which significantly improves efficiency model. Coordinate Attention (CA) mechanism added backbone network enhance model's ability localize extract features from regions. neck lightweight Content-Aware ReAssembly FEatures (CARAFE) upsampling operator capture retain detailed at low levels. Adam optimizer learning complex parameter spaces, thus improving performance.The experimental results showed that Params, FLOPs, Weight Size YOLOv8-Peas were 1.17M, 3.2G, 2.7MB, respectively, decreased 61.2%, 61%, 56.5% compared with YOLOv8-n. mAP was on par YOLOv8-n, reaching 98.7%, achieved speed 116.2FPS. PEG6000 simulate different drought environments analyze quantify genotypes peas, screened best drought-resistant varieties.Our effectively reduces deployment costs, efficiency, provides scientific theoretical basis genotype screening pea.

Язык: Английский

---

YOLOv5-ASFF: A Multistage Strawberry Detection Algorithm Based on Improved YOLOv5

Agronomy, Год журнала: 2023, Номер 13(7), С. 1901 - 1901

Опубликована: Июль 19, 2023

The smart farm is currently a hot topic in the agricultural industry. Due to complex field environment, intelligent monitoring model applicable this environment requires high hardware performance, and there are difficulties realizing real-time detection of ripe strawberries on small automatic picking robot, etc. This research proposes multistage strawberry algorithm YOLOv5-ASFF based improved YOLOv5. Through introduction ASFF (adaptive spatial feature fusion) module into YOLOv5, network can adaptively learn fused weights maps at each scale as way fully obtain image information strawberries. To verify superiority availability YOLOv5-ASFF, dataset containing variety scenarios, including leaf shading, overlapping fruit, dense was constructed experiment. method achieved 91.86% 88.03% for mAP F1, respectively, 98.77% AP mature-stage strawberries, showing strong robustness generalization ability, better than SSD, YOLOv3, YOLOv4, YOLOv5s. overcome influence environments improve under distribution shading conditions, provide technical support yield estimation harvest planning management.

Язык: Английский

---

Real-time and lightweight detection of grape diseases based on Fusion Transformer YOLO

Frontiers in Plant Science, Год журнала: 2024, Номер 15

Опубликована: Фев. 23, 2024

Introduction Grapes are prone to various diseases throughout their growth cycle, and the failure promptly control these can result in reduced production even complete crop failure. Therefore, effective disease is essential for maximizing grape yield. Accurate identification plays a crucial role this process. In paper, we proposed real-time lightweight detection model called Fusion Transformer YOLO 4 detection. The primary source of dataset comprises RGB images acquired from plantations situated North China. Methods Firstly, introduce high-performance VoVNet, which utilizes ghost convolutions learnable downsampling layer. This backbone further improved by integrating squeeze excitation blocks residual connections OSA module. These enhancements contribute accuracy while maintaining network. Secondly, an dual-flow PAN+FPN structure with Real-time adopted neck component, incorporating 2D position embedding single-scale Encoder into last feature map. modification enables performance detecting small targets. Finally, adopt Decoupled Head based on Task Aligned Predictor head balances speed. Results Experimental results demonstrate that FTR-YOLO achieves high across evaluation metrics, mean Average Precision (mAP) 90.67%, Frames Per Second (FPS) 44, parameter size 24.5M. Conclusion presented paper provides solution diseases. effectively assists farmers

Язык: Английский

---

Application of improved YOLOv7-based sugarcane stem node recognition algorithm in complex environments

Frontiers in Plant Science, Год журнала: 2023, Номер 14

Опубликована: Авг. 23, 2023

Introduction Sugarcane stem node detection is one of the key functions a small intelligent sugarcane harvesting robot, but accuracy severely degraded in complex field environments when shadow confusing backgrounds and other objects. Methods To address problem low arise environments, this paper proposes an improved model based on YOLOv7. First, SimAM (A Simple Parameter-Free Attention Module for Convolutional Neural Networks) attention mechanism added to solve feature loss due image global context information convolution process, which improves case blurring; Second, Deformable Network used replace some traditional layers original Finally, new bounding box regression function WIoU Loss introduced unbalanced sample quality, improve robustness generalization ability, accelerate convergence speed network. Results The experimental results show that mAP algorithm 94.53% F1 value 92.41, are 3.43% 2.21 respectively compared with YOLOv7 model, SOTA method 94.1%, improvement 0.43% achieved, effectively performance target model. Discussion This study provides theoretical basis technical support development may also provide reference types crops similar environments.

Язык: Английский

---

Compliant picking control of dragon fruit picking robot based on adaptive variable impedance

Biosystems Engineering, Год журнала: 2025, Номер 252, С. 126 - 143

Опубликована: Март 13, 2025

Язык: Английский

---

Lightweight highland barley detection based on improved YOLOv5

Plant Methods, Год журнала: 2025, Номер 21(1)

Опубликована: Март 24, 2025

Accurate and efficient assessment of highland barley (Hordeum vulgare L.) density is crucial for optimizing cultivation management practices. However, challenges such as overlapping spikes in unmanned aerial vehicle (UAV) images the computational requirements high-resolution image analysis hinder real-time detection capabilities. To address these issues, this study proposes an improved lightweight YOLOv5 model spike detection. We chose depthwise separable convolution (DSConv) ghost (GhostConv) backbone neck networks, respectively, to reduce parameter complexity. In addition, integration convolutional block attention module (CBAM) enhances model's ability focus on target object complex backgrounds. The results show that has a significant improvement performance. Precision recall increased by 3.1% 92.2% 86.2%, with F1 score 0.892. $$\hbox {AP}_{0.5}$$ reaches 92.7% 93.5% growth maturation stages, overall {mAP}_{0.5}$$ 93.1%. Compared baseline YOLOv5n model, number parameters floating-point operations (FLOPs) were reduced 70.6% 75.6%, enabling deployment without compromising accuracy. addition,the proposed outperformed mainstream algorithms Faster R-CNN, Mask RetinaNet, YOLOv7, YOLOv8, terms accuracy efficiency. Although also suffers from limitations insufficient generalization under varying lighting conditions reliance rectangular annotations, it provides valuable support reference development systems, which can help improve agricultural management.

Язык: Английский

---

Lightweight-Improved YOLOv5s Model for Grape Fruit and Stem Recognition

Agriculture, Год журнала: 2024, Номер 14(5), С. 774 - 774

Опубликована: Май 17, 2024

Mechanized harvesting is the key technology to solving high cost and low efficiency of manual harvesting, realizing mechanized lies in accurate fast identification localization targets. In this paper, a lightweight YOLOv5s model improved for efficiently identifying grape fruits stems. On one hand, it improves CSP module using Ghost module, reducing parameters through ghost feature maps cost-effective linear operations. other replaces traditional convolutions with deep further reduce model’s computational load. The trained on datasets under different environments (normal light, strong noise) enhance generalization robustness. applied recognition stems, experimental results show that overall accuracy, recall rate, mAP, F1 score are 96.8%, 97.7%, 98.6%, 97.2% respectively. average detection time GPU 4.5 ms, frame rate 221 FPS, weight size generated during training 5.8 MB. Compared original YOLOv5s, YOLOv5m, YOLOv5l, YOLOv5x models specific orchard environment greenhouse, proposed accuracy by 1%, decreases 0.2%, increases 0.4%, maintains same mAP. terms size, reduced 61.1% compared model, only 1.8% 5.5% Faster-RCNN SSD models, FPS increased 43.5% 11.05 times 8.84 CPU, 23.9 41.9 representing 31% improvement over model. test demonstrate lightweight-improved study, while maintaining significantly reduces enhances speed, can provide robotic harvesting.

Язык: Английский

---