An improved ShuffleNetV2 method based on ensemble self-distillation for tomato leaf diseases recognition

Frontiers in Plant Science, Год журнала: 2025, Номер 15

Опубликована: Янв. 21, 2025

Introduction Timely and accurate recognition of tomato diseases is crucial for improving yield. While large deep learning models can achieve high-precision disease recognition, these often have a number parameters, making them difficult to deploy on edge devices. To address this issue, study proposes an ensemble self-distillation method applies it the lightweight model ShuffleNetV2. Methods Specifically, based architecture ShuffleNetV2, multiple shallow at different depths are constructed establish distillation framework. Based fused feature map that integrates intermediate maps ShuffleNetV2 models, depthwise separable convolution layer introduced further extract more effective information. This ensures features from each fully preserved model, thereby overall performance model. The acting as teacher, dynamically transfers knowledge during training, significantly enhancing without changing original structure. Results Experimental results show optimized achieves accuracy 95.08%, precision 94.58%, recall 94.55%, F1 score 94.54% test set, surpassing such VGG16 ResNet18. Among has smallest parameter count highest accuracy. Discussion demonstrate suitable deployment devices real-time detection. Additionally, varying degrees compression providing flexibility deployment.

Язык: Английский

---

A multi-modal deep learning solution for precise pneumonia diagnosis: the PneumoFusion-Net model

Frontiers in Physiology, Год журнала: 2025, Номер 16

Опубликована: Март 12, 2025

Background Pneumonia is considered one of the most important causes morbidity and mortality in world. Bacterial viral pneumonia share many similar clinical features, thus making diagnosis a challenging task. Traditional diagnostic method developments mainly rely on radiological imaging require certain degree consulting experience, which can be inefficient inconsistent. Deep learning for classification multiple modalities, especially integrating data, has not been well explored. Methods The study introduce PneumoFusion-Net, deep learning-based multimodal framework that incorporates CT images, text, numerical lab test results, radiology reports improved diagnosis. In experiments, dataset 10,095 images was used-including associated data-most used training validation while keeping part it held-out set. Five-fold cross-validation order to evaluate this model, calculating different metrics including accuracy F1-Score. Results achieved 98.96% with 98% F1-score set, highly effective distinguishing bacterial from types pneumonia. This beneficial diagnosis, reducing misdiagnosis further improving homogeneity across various data sets patients. Conclusion PneumoFusion-Net offers an efficient approach by diverse sources, resulting high accuracy. Its potential integration could significantly reduce burden providing radiologists clinicians robust, automated tool.

Язык: Английский

---

CAPNet: tomato leaf disease detection network based on adaptive feature fusion and convolutional enhancement

Multimedia Systems, Год журнала: 2025, Номер 31(2)

Опубликована: Апрель 1, 2025

Язык: Английский

---

TomaFDNet: A multiscale focused diffusion-based model for tomato disease detection

Frontiers in Plant Science, Год журнала: 2025, Номер 16

Опубликована: Апрель 24, 2025

Tomatoes are one of the most economically significant crops worldwide, with their yield and quality heavily impacted by foliar diseases. Effective detection these diseases is essential for enhancing agricultural productivity mitigating economic losses. Current tomato leaf disease methods, however, encounter challenges in extracting multi-scale features, identifying small targets, complex background interference. The model Tomato Focus-Diffusion Network (TomaFDNet) was proposed to solve above problems. utilizes a focus-diffusion network (MSFDNet) alongside an efficient parallel convolutional module (EPMSC) significantly enhance extraction features. This combination particularly strengthens model's capability detect targets amidst backgrounds. Experimental results show that TomaFDNet reaches mean average precision (mAP) 83.1% detecting Early_blight, Late_blight, Leaf_Mold on leaves, outperforming classical object algorithms, including Faster R-CNN (mAP = 68.2%) You Only Look Once (YOLO) series (v5: mAP 75.5%, v7: 78.3%, v8: 78.9%, v9: 79%, v10: 77.5%, v11: 79.2%). Compared baseline YOLOv8 model, achieves 4.2% improvement mAP, which statistically (P < 0.01). These findings indicate offers valid solution precise

Язык: Английский

---

Multifactorial Tomato Leaf Disease Detection Based on Improved YOLOV5

Symmetry, Год журнала: 2024, Номер 16(6), С. 723 - 723

Опубликована: Июнь 11, 2024

Target detection algorithms can greatly improve the efficiency of tomato leaf disease and play an important technical role in intelligent cultivation. However, there are some challenges process, such as diversity complex backgrounds loss symmetry due to shadowing, existing methods have disadvantages terms deteriorating generalization ability insufficient accuracy. Aiming at above issues, a target model for based on deep learning with global attention mechanism, TDGA, is proposed this paper. The main idea TDGA includes three aspects. Firstly, adds mechanism (GAM) after up-sampling down-sampling, well SPPF module, feature extraction object, effectively reducing interference invalid targets. Secondly, uses switchable atrous convolution (SAConv) C3 module model’s detect. Thirdly, adopts efficient IoU (EIoU) instead complete (CIoU) solve ambiguous definition aspect ratio sample imbalance. In addition, influences different environmental factors single leaf, multiple leaves, shadows performance extensively experimented analyzed paper, which also verified robustness TDGA. experimental results show that average accuracy reaches 91.40%, 2.93% higher than original YOLOv5 network, YOLOv5, YOLOv7, YOLOHC, YOLOv8, SSD, Faster R-CNN, RetinaNet other networks, so be utilized more efficiently accurately, even environments.

Язык: Английский

---

Evaluation Model of Rice Seedling Production Line Seeding Quality Based on Deep Learning

Applied Sciences, Год журнала: 2024, Номер 14(7), С. 3098 - 3098

Опубликована: Апрель 7, 2024

A critical precondition for realizing mechanized transplantation in rice cultivation is the implementation of seedling tray techniques. To augment efficacy seeding, a precise evaluation quality these trays imperative. This research centers on analysis images, employing deep learning as foundational technology. The aim to construct computational model capable autonomously evaluating seeding within ambit intelligent processes. study proposes virtual grid-based image segmentation preprocessing method. It involves dividing complete into several grid images. These images are then classified and marked using an improved ResNet50 that integrates SE attention mechanism with Adam optimizer. Finally, objective detecting missing areas achieved by reassembling experimental results demonstrate model, integrating initial rate 0.01 over 50 iterations, attains test set accuracy 95.82%. surpasses AlexNet, DenseNet, VGG16 models respective margins 4.55%, 2.07%, 2.62%. introduces innovative automatic assessment quality. rapidly during phase; precisely identifying locations seeds individual trays; effectively calculating seed each tray. Such precision instrumental optimizing processes

Язык: Английский

---

Identification of tomato leaf diseases based on DGP-SNNet

Crop Protection, Год журнала: 2024, Номер 187, С. 106975 - 106975

Опубликована: Окт. 5, 2024

Язык: Английский

---

Revolutionizing Tomato Agriculture: Leaf Disease Detection Using CNN and Its Variants

2022 IEEE 7th International conference for Convergence in Technology (I2CT), Год журнала: 2024, Номер unknown

Опубликована: Апрель 5, 2024

India, being a large agricultural market is considered to be one of the major producers tomatoes in world having high economic value. However, volume and quality tomato crop output are diminishing day by owing several factors that impact productivity resulting significant losses for farmers. The escalating challenges agriculture, therefore, demand innovative solutions timely accurate identification plant diseases. Numerous works have been provided literature address these problems, however attaining accuracy still challenge. To tackle inadequacies, we proposed an analytical approach based on Convolutional Neural Networks with emphasis ResNet50 VGG16 architectures enable better complex disease pattern detection. For our experiment analysis evaluation, exploited labeled dataset obtained from Kaggle comprising 10,388 images 10 different leaf classes. Our illustrated satisfactory detection accuracy. It also outperforms some methods attains 99.63% 94.48% training at 20 epochs models respectively. enhanced result evident knowledge transfer learning imports pre-trained Resnet50 data augmentation techniques.

Язык: Английский

---

Ensemble of Visual Transformer and Deep Neural Networks for Recognizing Sunflower Diseases from Photographs

Lecture notes in networks and systems, Год журнала: 2024, Номер unknown, С. 15 - 24

Опубликована: Янв. 1, 2024

Язык: Английский

---