MPG-YOLO: Enoki Mushroom Precision Grasping with Segmentation and Pulse Mapping

Agronomy, Journal Year: 2025, Volume and Issue: 15(2), P. 432 - 432

Published: Feb. 10, 2025

The flatness of the cut surface in enoki mushrooms (Flammulina filiformis Z.W. Ge, X.B. Liu & Zhu L. Yang) is a key factor quality classification. However, conventional automatic cutting equipment struggles with deformation issues due to its inability adjust grasping force based on individual mushroom sizes. To address this, we propose an improved method that integrates visual feedback dynamically execution end, enhancing precision. Our approach enhances YOLOv8n-seg Star Net, SPPECAN (a reconstructed SPPF efficient channel attention), and C2fDStar (C2f Net deformable convolution) improve feature extraction while reducing computational complexity loss. Additionally, introduce mask ownership judgment merging optimization algorithm correct positional offsets, internal disconnections, boundary instabilities area predictions. Based optimize parameters using centroid-based region width measurement establish width-to-PWM mapping model for precise conversion from data gripper control. Experiments real-situation settings demonstrate effectiveness our method, achieving mean average precision (mAP50:95) 0.743 segmentation, 4.5% improvement over YOLOv8, detection speed 10.3 ms target error only 0.14%. proposed relationship enables adaptive end-effector control, resulting 96% success rate 98% qualified rate. These results confirm feasibility provide strong technical foundation intelligent automation systems.

Language: Английский

Determination of Optimal Dataset Characteristics for Improving YOLO Performance in Agricultural Object Detection

Agriculture, Journal Year: 2025, Volume and Issue: 15(7), P. 731 - 731

Published: March 28, 2025

Recent advances in artificial intelligence and computer vision have led to significant progress the use of agricultural technologies for yield prediction, pest detection, real-time monitoring plant conditions. However, collecting large-scale, high-quality image datasets agriculture sector remains challenging, particularly specialized such as disease images. This study analyzed effects size (320–640+) number labels on performance a YOLO-based object detection model using diverse strawberries, tomatoes, chilies, peppers. Model was evaluated intersection over union average precision (AP), where AP curve smoothed Savitzky–Golay filter EEM. The results revealed that increasing improved certain degree, after which gradually diminished. Furthermore, while from 320 640 substantially enhanced performance, additional increases beyond yielded only marginal improvements. training time graphics processing unit usage scaled linearly with sizes, larger images require greater computational resources. These findings underscore importance an optimal strategy selecting label quantity under resource constraints real-world development.

Language: Английский

YOLOv5s-BiPCNeXt, a Lightweight Model for Detecting Disease in Eggplant Leaves

Plants, Journal Year: 2024, Volume and Issue: 13(16), P. 2303 - 2303

Published: Aug. 19, 2024

Ensuring the healthy growth of eggplants requires precise detection leaf diseases, which can significantly boost yield and economic income. Improving efficiency plant disease identification in natural scenes is currently a crucial issue. This study aims to provide an efficient method suitable for scenes. A lightweight model, YOLOv5s-BiPCNeXt, proposed. model utilizes MobileNeXt backbone reduce network parameters computational complexity includes C3-BiPC neck module. Additionally, multi-scale cross-spatial attention mechanism (EMA) integrated into network, nearest neighbor interpolation algorithm replaced with content-aware feature recombination operator (CARAFE), enhancing model's ability perceive multidimensional information extract multiscale features improving spatial resolution map. These improvements enhance accuracy eggplant leaves, effectively reducing missed incorrect detections caused by complex backgrounds localization small lesions at early stages brown spot powdery mildew diseases. Experimental results show that YOLOv5s-BiPCNeXt achieves average precision (AP) 94.9% disease, 95.0% mildew, 99.5% leaves. Deployed on Jetson Orin Nano edge device, attains recognition speed 26 FPS (Frame Per Second), meeting real-time requirements. Compared other algorithms, demonstrates superior overall performance, accurately detecting diseases under conditions offering valuable technical support prevention treatment

Language: Английский