Memuat
Memuat
Build system tomato leaf disease detection with PyTorch: transfer learning MobileNetV2 & ResNet50, fine-tuning, then automated hyperparameter optimization with Optuna, until web application. The workflow follows real research so it can directly become a thesis, dissertation, or publication framework.
Swipe to view curriculum, architecture, and model results.
Test Accuracy MobileNetV2 + Optuna (10 classes)
Accuracy Boost from Feature Extraction to Fine-Tuning
Optuna Trial with TPE Sampler & Median Pruner
Training time on CPU, no GPU
Every stage of AI agriculture research is practiced directly with PyTorch code.
2.000 photos of tomato leaves from 10 classes: healthy, bacterial spot, early and late blight, fungus, spider mites, virus, stratified split 70/15/15.
RandomResizedCrop, flip, rotation, ColorJitter, ImageNet normalization, and in-memory dataset for fast training.
MobileNetV2 and ResNet50 pretrained ImageNet from torchvision: replace classification head and freeze model body.
Opening upper block, small learning rate, and cosine annealing, accuracy rises from 76% to 93%.
Objective function, TPE sampler, median pruner, and hyperparameter importance for finding the best configuration.
Precision–recall–F1 per class, confusion matrix, prediction error analysis, and Streamlit disease detection application.
Video 7–15 minutes per module, complete with PDF guide. Module 1 can be watched for free.
Complete AI agriculture research framework: baseline model, proposed model, optimization, and evaluation ready for chapter 4.
Transfer learning pipeline easily adaptable to different crops, diseases, or field datasets.
Understand how to build a plant disease diagnosis tool from leaf photos.
Practice PyTorch, torchvision, Optuna, and Streamlit from data to application.
We help with model implementation, experiments, evaluation, and result writing for theses, dissertations, or publications.
Lihat Jasa Riset AI PertanianA technique that uses a model already trained with millions of images (ImageNet) and then adapts it for a new task, so that it only needs a small amount of data, in this course, 140 training images per class.
No. 128x120 pixel images make feature extraction, fine-tuning, and Optuna optimization each complete in a few minutes on a CPU.
PlantVillage (Hughes & Salathé, 2015, license CC BY-SA 3.0), 10-class subset of tomato leaves. This dataset is widely used in publications and can be used with proper citation.
Yes. Premium Course subscribers and class buyers get a zip file containing the dataset, all source code, Optuna results, trained model, and Streamlit application.
Yes. Replace the contents of the dataset folder with images of other plants (potato, rice, chili, etc.) and the same pipeline can be used directly.
Yes, live online classes and AI agricultural research mentoring services are available for thesis, dissertation, or publication.
Consult your research topic, we help you choose the most suitable learning package.
Free Consultation