Research on agricultural disease recognition methods based on very large Kernel convolutional network-RepLKNet DOI Creative Commons

Guoquan Pei,

Xueying Qian,

B. Zhou

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 15, 2025

Agricultural diseases pose significant challenges to plant production. With the rapid advancement of deep learning, accuracy and efficiency disease identification have substantially improved. However, conventional convolutional neural networks that rely on multi-layer small-kernel structures are limited in capturing long-range dependencies global contextual information due their constrained receptive fields. To overcome these limitations, this study proposes a recognition method based RepLKNet, architecture with large kernel designs significantly expand field enhance feature representation. Transfer learning is incorporated further improve training model performance. Experiments conducted Plant Diseases Training Dataset, comprising 95,865 images across 61 categories, demonstrate effectiveness proposed method. Under five-fold cross-validation, achieved an overall (OA) 96.03%, average (AA) 94.78%, Kappa coefficient 95.86%. Compared ResNet50 (OA: 95.62%) GoogleNet 94.98%), demonstrates competitive or superior Ablation experiments reveal replacing kernels 3×3 5×5 convolutions results reductions up 1.1% OA 1.3% AA, confirming design. These robustness capability RepLKNet tasks.

Language: Английский

Optimizing Pear Leaf Disease Detection Through PL-DenseNet DOI
Yonis Gulzar, Zeynep Ünal

Deleted Journal, Journal Year: 2025, Volume and Issue: 67(1)

Published: Feb. 1, 2025

Language: Английский

Citations

0
Mushroom image classification and recognition based on improved ConvNeXt V2 DOI
Shulong Zhang,

Kexin Zhao,

Yukang Huo

et al.

Journal of Food Science, Journal Year: 2025, Volume and Issue: 90(3)

Published: March 1, 2025

Abstract Using on‐site images to classify and identify wild mushroom species is the most effective way prevent incidents of harm caused by eating mushrooms. However, complexity natural scenes similarity morphology bring challenges for accurate classification recognition. To this end, paper proposes an improved ConvNeXt V2 network model recognition mushrooms in complex similar appearances. First, study applies data enhancement techniques such as image flipping, adding noise mosaic solve problem dataset equalization, constructs a containing 18 categories number 10,986 images. Second, cross‐modular approach used extract fuse features different dimensions enhance feature capture capability model. In addition, optimized one‐hot coding spatial pyramid pooling techniques. The experimental results show that outperforms comparative models ResNet, MobileVit, Swin Transformer, ConvNeXt, terms accuracy, precision, recall, F1‐Score, which are 96.7%, 96.84%, 96.83%, 96.84%. ablation experiments further verify effectiveness superiority proposed improvement strategy enhancing performance, can effectively improve efficiency accuracy Practical Application : be identification edible nonedible mushroom, it provide technical support reduce incidence poisoning ensure food safety.

Language: Английский

Citations

0
MoSViT: a lightweight vision transformer framework for efficient disease detection via precision attention mechanism DOI Creative Commons
Yuanqi Chen, Aiping Wang, Ziyang Liu

et al.

Frontiers in Artificial Intelligence, Journal Year: 2025, Volume and Issue: 8

Published: March 26, 2025

Maize, a globally essential staple crop, suffers significant yield losses due to diseases. Traditional diagnostic methods are often inefficient and subjective, posing challenges for timely accurate pest management. This study introduces MoSViT, an innovative classification model leveraging advanced machine learning computer vision technologies. Built on the MobileViT V2 framework, MoSViT integrates CLA focus mechanism, DRB module, Block, LeakyRelu6 activation function enhance feature extraction accuracy while reducing computational complexity. Trained dataset of 3,850 images encompassing Blight, Common Rust, Gray Leaf Spot, Healthy conditions, achieves exceptional performance, with accuracy, Precision, Recall, F1 Score 98.75%, 98.73%, 98.72%, respectively. These results surpass leading models such as Swin Transformer V2, DenseNet121, EfficientNet in both parameter efficiency. Additionally, model's interpretability is enhanced through heatmap analysis, providing insights into its decision-making process. Testing small sample datasets further demonstrates MoSViT's generalization capability potential small-sample detection scenarios.

Language: Английский

Citations

0
Comparison of deep learning models in automatic classification of coffee bean species DOI Creative Commons
Adem Korkmaz, Tarık Talan, Selahattin Koşunalp

et al.

PeerJ Computer Science, Journal Year: 2025, Volume and Issue: 11, P. e2759 - e2759

Published: April 7, 2025

As one of the most widely consumed beverages worldwide, coffee is characterized by its diverse flavor profiles and complex production processes. In this study, deep learning-based image processing techniques are employed for automatic classification bean species with high accuracy. To achieve this, images three different (Starbucks Pike Place, Espresso, Kenya) were classified using five CNN-based models: Xception, DenseNet201, InceptionV3, InceptionResNetV2, DenseNet121. The dataset comprises 1,554 images. Cross-validation was applied to assess models’ performance, accuracy evaluated performance metrics. Among tested models, InceptionV3 achieved highest (93%) precision (95%), lowest loss rate (0.12), making it effective model in study. a result experiments, average success rates models determined as follows: 93% 92% DenseNet121, 91% 90% DenseNet201. These findings indicate that demonstrates performance. It anticipated study will make significant contributions applications classification.

Language: Английский

Citations

0
Time-Sensitive Bruise Detection in Plums Using PlmNet with Transfer Learning DOI Open Access
Yonis Gulzar, Zeynep Ünal

Procedia Computer Science, Journal Year: 2025, Volume and Issue: 257, P. 127 - 132

Published: Jan. 1, 2025

Language: Английский

Citations

0
Research on agricultural disease recognition methods based on very large Kernel convolutional network-RepLKNet DOI Creative Commons

Guoquan Pei,

Xueying Qian,

B. Zhou

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 15, 2025

Agricultural diseases pose significant challenges to plant production. With the rapid advancement of deep learning, accuracy and efficiency disease identification have substantially improved. However, conventional convolutional neural networks that rely on multi-layer small-kernel structures are limited in capturing long-range dependencies global contextual information due their constrained receptive fields. To overcome these limitations, this study proposes a recognition method based RepLKNet, architecture with large kernel designs significantly expand field enhance feature representation. Transfer learning is incorporated further improve training model performance. Experiments conducted Plant Diseases Training Dataset, comprising 95,865 images across 61 categories, demonstrate effectiveness proposed method. Under five-fold cross-validation, achieved an overall (OA) 96.03%, average (AA) 94.78%, Kappa coefficient 95.86%. Compared ResNet50 (OA: 95.62%) GoogleNet 94.98%), demonstrates competitive or superior Ablation experiments reveal replacing kernels 3×3 5×5 convolutions results reductions up 1.1% OA 1.3% AA, confirming design. These robustness capability RepLKNet tasks.

Language: Английский

Citations

0