Ripeness Classification of Cavendish Bananas using Multi-object Detection Approach DOI
Dito Eka Cahya, Zaid Cahya, Heru Taufiqurrohman

et al.

Published: Nov. 15, 2023

Cavendish bananas, recognized for their global economic and nutritional value, often pose challenges in ensuring consistent ripeness quality throughout the supply chain, which is essential efficient post-harvest management optimal consumer experience. Traditional methods, reliant on human visual inspection, are subjective inconsistent. This paper proposes a novel multi-object detection approach accurately classifying of multiple bananas an image using computer vision deep learning techniques. We employed Yolov5 model, state-of-the-art object architecture standardized classification system to simultaneously identify categorize every banana image. To train diverse dataset 600 bunches at different stages was collected, augmented, annotated with ground truth labels. Training result shows that network successfully delineates input images while predicting classes. The proposed achieves 98.8% mean average precision, 90.5% 92.6% recall, even when test contain overlapping background colors. compact size made it applicable embedded system, enabling realtime simple uploads from handheld devices. research contributes advancement agricultural technology opens avenues future studies fruit analysis food assessment.

Language: Английский

Detection and Counting of Small Target Apples under Complicated Environments by Using Improved YOLOv7-tiny DOI Creative Commons
Li Ma, Liya Zhao, Zixuan Wang

et al.

Agronomy, Journal Year: 2023, Volume and Issue: 13(5), P. 1419 - 1419

Published: May 20, 2023

Weather disturbances, difficult backgrounds, the shading of fruit and foliage, other elements can significantly affect automated yield estimation picking in small target apple orchards natural settings. This study uses MinneApple public dataset, which is processed to construct a dataset 829 images with complex weather, including 232 fog scenarios 236 rain scenarios, proposes lightweight detection algorithm based on upgraded YOLOv7-tiny. In this study, backbone network was constructed by adding skip connections shallow features, using P2BiFPN for multi-scale feature fusion reuse at neck, incorporating ULSAM attention mechanism reduce loss focusing correct discard redundant thereby improving accuracy. The experimental results demonstrate that model has an mAP 80.4% rate 0.0316. 5.5% higher than original model, size reduced 15.81%, reducing requirement equipment; terms counts, MAE RMSE are 2.737 4.220, respectively, 5.69% 8.97% lower model. Because its improved performance stronger robustness, offers fresh perspectives hardware deployment orchard estimation.

Language: Английский

Citations

52
Advances in apple’s automated orchard equipment: A comprehensive research DOI
Mustafa Mhamed, Zhao Zhang, Jiangfan Yu

et al.

Computers and Electronics in Agriculture, Journal Year: 2024, Volume and Issue: 221, P. 108926 - 108926

Published: April 23, 2024

Language: Английский

Citations

20
O2RNet: Occluder-occludee relational network for robust apple detection in clustered orchard environments DOI Creative Commons
Pengyu Chu, Zhaojian Li, Kaixiang Zhang

et al.

Smart Agricultural Technology, Journal Year: 2023, Volume and Issue: 5, P. 100284 - 100284

Published: July 11, 2023

Automated apple harvesting has attracted significant research interest in recent years because of its great potential to address the issues labor shortage and rising costs. One key challenge automated is accurate robust detection, due complex orchard environments that involve varying lighting conditions, fruit clustering foliage/branch occlusions. Apples are often grown clusters on trees, which may be mis-identified as a single thus causes localization for subsequent robotic operations. In this paper, we present development novel deep learning-based detection framework, called Occluder-Occludee Relational Network (O2RNet), apples clustered situations. A comprehensive dataset RGB images were collected two varieties under different conditions (overcast, direct lighting, back lighting) with degrees occlusions, annotated made available public. occlusion-aware network was developed feature expansion structure incorporated into convolutional neural networks extract additional features generated by original occluded apples. Comprehensive evaluations O2RNet performed using images, outperformed 12 other state-of-the-art models higher accuracy 94% F1-score 0.88 detection. provides an enhanced method apples, critical harvesting.

Language: Английский

Citations

17
Object Detection Algorithm of Transmission Lines Based on Improved YOLOv5 Framework DOI Creative Commons
Hao Zhang, Xianjun Zhou, Yike Shi

et al.

Journal of Sensors, Journal Year: 2024, Volume and Issue: 2024, P. 1 - 14

Published: Jan. 9, 2024

Foreign objects easily attach to the transmission lines because of various laying methods and complex, changing environment. They have a significant impact on safe operation capability if these foreign are not detected removed in time. An improved YOLOv5 technique is provided detect due low-foreign object recognition accuracy image detection. The method first reduces computation memory consumption by introducing RepConv structure, further improves detection speed model embedding C2F structure. This finally optimized neural network Meta-ACON activation function. results indicate that average can reach 96.9%, which 2.2% higher than before. Additionally, corresponding 258.36 frames/second, surpasses existing mainstream target models, performing better terms balance inference accuracy. Consequently, effectiveness superiority algorithm been proved.

Language: Английский

Citations

4
Classification and Forecasting of Water Stress in Tomato Plants Using Bioristor Data DOI Creative Commons
Manuele Bettelli,

Filippο Vurro,

Riccardo Pecori

et al.

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 34795 - 34807

Published: Jan. 1, 2023

Water stress and in particular drought are some of the most significant factors affecting plant growth, food production, thus security. Furthermore, possibility to predict shape irrigation on real demands is priceless. The objective this study characterize, classify, forecast water tomato plants by means vivo time data obtained through a novel sensor, named bioristor, different artificial intelligence models. First all, we have applied classification models, namely Decision Trees Random Forest, try distinguish four statuses plants. Then predicted, help recurrent neural networks, future status when considering both binary (water stressed not stressed) four-status scenario. results very good terms accuracy, precision, recall, f-measure, resulting confusion matrices, they suggest that considered features coming from together with used AI can be successfully applied, future, real-world on-the-field smart

Language: Английский

Citations

10
Crack-Detection Method for Asphalt Pavement Based on the Improved YOLOv5 DOI

Gangting Tang,

Chao Yin,

Xixuan Zhang

et al.

Journal of Performance of Constructed Facilities, Journal Year: 2024, Volume and Issue: 38(2)

Published: Feb. 8, 2024

In view of the low identification accuracy crack-detection technology asphalt pavement under current conditions complex (subject to strong light, water on road, debris, and so on), an algorithm based improved YOLOv5s was proposed by building data set for cracks. The first step make following improvements original model according characteristics crack set: k-means++ used recluster anchor points set, initial frame matching obtained replace default in YOLOv5 model. prediction part model, Convolutional Block Attention Module (CBAM) added order channel then space improve detection ability small CIoU_Loss function as regression loss GIoU_Loss positioning frame. second perform ablation experiment This would prove that each improvement scheme could increase without conflict. final compare with various classic target models this paper: Crack Forest Data (CFD), Crack500 Crack200 set. results showed better than other models. [email protected] mAP@[0.5:0.95] paper were 90.58% 56.08%, respectively, which much higher These findings indicate had provide a theoretical basis automatic

Language: Английский

Citations

2
Study on Different Apple Ripeness Detection Based on Improved YOLOv5 DOI

YuluCai,

Fangchu Wanghan,

Anwen Shen

et al.

Communications in computer and information science, Journal Year: 2024, Volume and Issue: unknown, P. 177 - 188

Published: Dec. 21, 2024

Language: Английский

Citations

0
Research on Red Jujubes Recognition Based on a Convolutional Neural Network DOI Creative Commons
Jingming Wu, Cuiyun Wu,

Huaying Guo

et al.

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(11), P. 6381 - 6381

Published: May 23, 2023

Red jujube is one of the most important crops in China. In order to meet needs scientific and technological development industry, solve problem poverty, realize backward advantage, promote economic development, smart agriculture essential. The main objective this study was conduct an online detection unpicked red jujubes detect as many picture possible while minimizing occurrence overfitting underfitting. Experiments were conducted using Histogram Oriented Gradients + Support Vector Machine (HOG+SVM) traditional method You Only Look Once version 5 (YOLOV5) Faster R-CNN modern deep learning methods. precision, recall, F1 score compared obtain a better algorithm. also introduced AlexNet model with attempting combine it other algorithms maximize accuracy. Labeling used label training images YOLOV5 Regions CNN Features (Faster R-CNN) train machine so that computer recognized these features when saw new unlabeled data subsequent experiments. experimental results show recognition jujubes, performed than HOG SVM algorithm, which presents values 93.55%, 82.79%, 87.84% respectively; although algorithm relatively quicker perform. precision obviously more efficiency study, experiments, had 100% 99.65% 99.82%, 83% non-underfitting for images, all higher YOLOV5′s values, 97.17% 98.56%, 64.42% non-underfitting. therefore, works best.

Language: Английский

Citations

1
A Social Distancing Detection Method for Public Area Crowds Based on YOLO DOI

冰 陈

Artificial Intelligence and Robotics Research, Journal Year: 2023, Volume and Issue: 12(03), P. 199 - 208

Published: Jan. 1, 2023

Language: Английский

Citations

0
Retracted: Fast Recognition Method for Multiple Apple Targets in Complex Occlusion Environment Based on Improved YOLOv5 DOI Creative Commons

Journal of Sensors

Journal of Sensors, Journal Year: 2023, Volume and Issue: 2023(1)

Published: Jan. 1, 2023

Language: Английский

Citations

0