YOLOv5-ACS: Improved Model for Apple Detection and Positioning in Apple Forests in Complex Scenes

Forests, Год журнала: 2023, Номер 14(12), С. 2304 - 2304

Опубликована: Ноя. 24, 2023

Apple orchards, as an important center of economic activity in forestry special crops, can achieve yield prediction and automated harvesting by detecting locating apples. Small apples, occlusion, dim lighting at night, blurriness, cluttered backgrounds, other complex scenes significantly affect the automatic estimation To address these issues, this study proposes apple detection algorithm, “YOLOv5-ACS (Apple Complex Scenes)”, based on YOLOv5s. Firstly, space-to-depth-conv module is introduced to avoid information loss, a squeeze-and-excitation block added C3 learn more information. Secondly, context augmentation incorporated enrich feature pyramid network. By combining shallow features backbone P2, low-level object are retained. Finally, addition aggregation CoordConv aggregates spatial pixel pixel, perceives map, enhances semantic global perceptual ability object. We conducted comparative tests various scenarios validated robustness YOLOv5-ACS. The method achieved 98.3% 74.3% for [email protected] [email protected]:0.95, respectively, demonstrating excellent capabilities. This paper creates scene dataset apples trees designs improved model, which provide accurate recognition positioning robots improve production efficiency.

Язык: Английский

---

S³AD: Semi-supervised Small Apple Detection in Orchard Environments

2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), Год журнала: 2024, Номер unknown, С. 7061 - 7070

Опубликована: Янв. 3, 2024

Crop detection is integral for precision agriculture applications such as automated yield estimation or fruit picking. However, crop detection, e.g., apple in orchard environments remains challenging due to a lack of large-scale datasets and the small relative size crops image. In this work, we address these challenges by reformulating task semi-supervised manner. To end, provide large, high-resolution dataset MAD ¹ comprising 105 labeled images with 14,667 annotated instances 4,440 unlabeled images. Utilizing dataset, also propose novel Semi-Supervised Small Apple Detection system S xmlns:xlink="http://www.w3.org/1999/xlink">3 AD based on contextual attention selective tiling improve apples, while limiting computational overhead. We conduct an extensive evaluation MSU showing that substantially outperforms strong fully-supervised baselines, including several object systems, up 14.9%. Additionally, exploit detailed annotations our w.r.t. properties analyze influence level occlusion results various quantifying current challenges.

Язык: Английский

---

Enhanced Real-Time Target Detection for Picking Robots Using Lightweight CenterNet in Complex Orchard Environments

Agriculture, Год журнала: 2024, Номер 14(7), С. 1059 - 1059

Опубликована: Июнь 30, 2024

The rapid development of artificial intelligence and remote sensing technologies is indispensable for modern agriculture. In orchard environments, challenges such as varying light conditions shading complicate the tasks intelligent picking robots. To enhance recognition accuracy efficiency apple-picking robots, this study aimed to achieve high detection in complex environments while reducing model computation time consumption. This utilized CenterNet neural network framework, introducing gray-centered RGB color space vertical decomposition maps employing grouped convolutions depth-separable design a lightweight feature extraction network, Light-Weight Net, comprising eight bottleneck structures. Based on results, 3D coordinates point were determined within camera coordinate system by using transformation relationship between image’s physical system, along with depth map distance information map. Experimental results obtained testbed an orchard-picking robot indicated that proposed achieved average precision (AP) 96.80% test set, real-time performance 18.91 frames per second (FPS) size only 17.56 MB. addition, root-mean-square error positioning was 4.405 mm, satisfying high-precision requirements vision environments.

Язык: Английский

---

Balancing Accuracy and Efficiency: The Status and Challenges of Agricultural Multi-Arm Harvesting Robot Research

Agronomy, Год журнала: 2024, Номер 14(10), С. 2209 - 2209

Опубликована: Сен. 25, 2024

As the global fruit growing area continues to increase and population aging problem intensify, vegetable production is constrained by difficulties of labor shortages high costs. Single-arm harvesting robots are inefficient, in order balance accuracy efficiency, research on multi-arm has become a hot topic. This paper summarizes performance indoor outdoor environments from aspects automatic navigation technology, identification localization, workspace optimization, task planning analyzes their advantages challenges practical applications. The results show that lack application field for robots, low rate non-structured environments, complexity algorithms robots’ main hindering wide-scale application. Future studies need focus building standardized environment control amount information acquired optimize strategy these challenges, which an important direction robots.

Язык: Английский

---

High-Precision Fruit Localization Using Active Laser-Camera Scanning: Robust Laser Line Extraction for 2D-3D Transformation

Опубликована: Янв. 1, 2023

Recent advancements in deep learning-based approaches have led to remarkable progress fruit detection, enabling robust identification complex environments. However, much less has been made on 3D localization, which is equally crucial for robotic harvesting. Complex shape/orientation, clustering, varying lighting conditions, and occlusions by leaves branches greatly restricted existing sensors from achieving accurate localization the natural orchard environment. In this paper, we report design of a novel technique, called Active Laser-Camera Scanning (ALACS), achieve localization. The ALACS hardware setup comprises red line laser, an RGB color camera, linear motion slide, external RGB-D camera. Leveraging principles dynamic-targeting laser-triangulation, enables precise transformation projected 2D laser surface apples positions. To facilitate pattern acquisitions, Laser Line Extraction (LLE) method proposed high-precision feature extraction apples. Comprehensive evaluations LLE demonstrated its ability extract patterns under variable occlusion conditions. system achieved average apple accuracies 6.9 11.2 mm at distances ranging 1.0 m 1.6 m, compared 21.5 commercial RealSense indoor experiment. Orchard that 95% detachment rate versus 71% By overcoming challenges research contributes advancement harvesting technology.

Язык: Английский

---

Enhancing horticultural harvest efficiency: The role of moisture content in ultrasonic cutting of tomato stems

Scientia Horticulturae, Год журнала: 2024, Номер 338, С. 113698 - 113698

Опубликована: Окт. 5, 2024

Язык: Английский

---

Multi-scale convolution and dynamic task interaction detection head for efficient lightweight plum detection

Food and Bioproducts Processing, Год журнала: 2024, Номер 149, С. 353 - 367

Опубликована: Дек. 11, 2024

Язык: Английский

---