Evaluation of YOLO Object Detectors for Weed Detection in Different Turfgrass Scenarios

Sportelli, Mino; Apolo-Apolo, O.E.; Fontanelli, Marco; Frasconi, Christian; Raffaelli, Michele; Peruzzi, Andrea; Ruiz, Manuel Pérez

doi:10.3390/app13148502

Cited by 23 publications

(14 citation statements)

References 50 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In automated weed control systems, actuators are constrained by the limited time available to process images and execute treatments. 55 Integrating these high-speed systems into robotic platforms facilitates real-time detection and action, even while the machinery is in motion. In this study, GoogLeNet and ResNet achieved notable speeds of 33.33 and 29.95 fps, respectively.…”

Section: Discussionmentioning

confidence: 99%

Detection and coverage estimation of purple nutsedge in turf with image classification neural networks

Jin,

Han,

Zhao

et al. 2024

Pest Management Science

View full text Add to dashboard Cite

BACKGROUNDAccurate detection of weeds and estimation of their coverage is crucial for implementing precision herbicide applications. Deep learning (DL) techniques are typically used for weed detection and coverage estimation by analyzing information at the pixel or individual plant level, which requires a substantial amount of annotated data for training. This study aims to evaluate the effectiveness of using image classification neural networks (NNs) for detecting and estimating weed coverage in bermudagrass turf.RESULTSThe weed detection NNs, including DenseNet, GoogLeNet, and ResNet, exhibited high overall accuracy and F1 scores (≥0.971) throughout the k‐fold cross‐validation. DenseNet outperformed GoogLeNet and ResNet with the highest overall accuracy and F1 scores (0.977). Among the evaluated NNs, the DenseNet showed the highest overall accuracy and F1 scores (0.996) in the validation and testing datasets for estimating weed coverage. The inference speed of ResNet was similar to GoogLeNet but noticeably faster than DenseNet. ResNet was the most efficient and accurate deep convolution neural network (DCNN) for weed detection and coverage estimation.CONCLUSIONThese results demonstrated that the developed NNs could effectively detect weeds and estimate their coverage in bermudagrass turf, allowing the calculation of herbicide requirements for variable‐rate herbicide applications. The proposed method can be employed in a machine vision‐based autonomous site‐specific spraying system of smart sprayers.This article is protected by copyright. All rights reserved.

show abstract

Section: Discussionmentioning

confidence: 99%

Detection and coverage estimation of purple nutsedge in turf with image classification neural networks

Jin,

Han,

Zhao

et al. 2024

Pest Management Science

View full text Add to dashboard Cite

show abstract

“…Pada penelitian membandingkan beberapa metode YOLO seperti YOLOv5m, YOLOv6l, YOLOv7, YOLOv8l, dengan EfficientDet. Hasil penelitian tersebut menunjukkan bahwa YOLOv8 memiliki performansi paling tinggi dengan nilai precision mencapai 0,9476 [17].…”

Section: A Pendahuluanunclassified

Implementasi Algoritma EfficientDet-D0 dan SSD-MobileNet-V2 FPNLite untuk Sistem Deteksi Gulma

Nailul Muna,

Norma Ningsih,

Nanang Syahroni

et al. 2024

ijcs

View full text Add to dashboard Cite

Gulma merupakan tumbuhan yang tumbuh disekitar tanaman dan dapat merugikan tanaman yang dibudidayakan. Pengendalian gulma menjadi faktor penting yang dapat mempengaruhi produktivitas tanaman. Pengendalian gulma dapat ditanggulangi dengan melakukan penyemprotan pestisida pada gulma. Cakupan penyemprotan yang tepat sasaran dapat dilakukan untuk mencegah timbulnya masalah limbah. Sistem pertanian cerdas sangat dibutuhkan untuk mengatasi permasalahan tersebut, seperti deteksi gulma yang memanfaatkan teknik deep learning. Pada penelitian ini membangun sistem deteksi gulma yamg mengimplementasikan EfficientDet-D0 dan SSD-MobileNet-V2 FPNLite. Dataset yang digunakan berjumlah 941 citra gulma yang kemudian dilakukan pelabelan untuk data latih dan data uji. Sistem menunjukkan kinerja yang baik untuk mendeteksi gulma dengan accuracy berturut-turut dari EfficientDet-D0 dan SSD-MobileNet-V2 FPNLite mencapai 95,69% dan 99,138%. Hasil tersebut menunjukkan EfficientDet-D0 dan SSD-MobileNet-V2 FPNLite dapat mendukung dalam pengendalian gulma.

show abstract

“…After matrix multiplication of the attention branch and the depth convolution branch, they are added to the original residual branch x to obtain the final output F 3 in Equation (10).…”

Section: Feature Fusion Network Improvementmentioning

confidence: 99%

“…In the context of coal mine drilling sites, these algorithms are unsuitable for real-time deployment of industrial equipment. On the other hand, first-stage algorithms like Single Shot MultiBox Detector(SSD) [9] and You Only Look Once(YOLO) series [10] exhibit fast detection speeds. In fact, in certain, specific domains, comparable detection accuracy can be achieved by these algorithms to that of two-stage algorithms.…”

Section: Introductionmentioning

confidence: 99%

Research on Automatic Counting of Drill Pipes for Underground Gas Drainage in Coal Mines Based on YOLOv7-GFCA Model

Chen,

Dong,

She

2023

Applied Sciences

View full text Add to dashboard Cite

Gas explosions threaten the safety of underground coal mining. Mining companies use drilling rigs to extract the gas to reduce its concentration. Drainage depth is a key indicator of gas drainage; accidents will be caused by going too deep. Since each drill pipe has the same length, the actual extraction depth is equivalent to the number of drill pipes multiplied by the length of a single drill pipe. Unnecessary labor is consumed and low precision is achieved by manual counting. Therefore, the drill pipe counting method of YOLOv7-GFCA target detection is proposed, and the counting is realized by detecting the movement trajectory of the drilling machine in the video. First, Lightweight GhostNetV2 is used as the feature extraction network of the model to improve the detection speed. Second, the (Fasternet-Coordinate-Attention) FCA network is fused into a feature fusion network, which improves the expression ability of the rig in complex backgrounds such as coal dust and strong light. Finally, Normalized Gaussian Wasserstein Distance (NWD) loss function is used to improve rig positioning accuracy. The experimental results show that the improved algorithm reaches 99.5%, the model parameters are reduced by 2.325 × 106, the weight file size is reduced by 17.8 M, and the detection speed reaches 80 frames per second. The movement trajectory of the drilling rig target can be accurately obtained by YOLOv7-GFCA, and the number of drill pipes can be obtained through coordinate signal filtering. The accuracy of drill pipe counting reaches 99.8%, thus verifying the feasibility and practicability of the method.

show abstract

Evaluation of YOLO Object Detectors for Weed Detection in Different Turfgrass Scenarios

Cited by 23 publications

References 50 publications

Detection and coverage estimation of purple nutsedge in turf with image classification neural networks

Detection and coverage estimation of purple nutsedge in turf with image classification neural networks

Implementasi Algoritma EfficientDet-D0 dan SSD-MobileNet-V2 FPNLite untuk Sistem Deteksi Gulma

Research on Automatic Counting of Drill Pipes for Underground Gas Drainage in Coal Mines Based on YOLOv7-GFCA Model

Contact Info

Product

Resources

About