---

Aim of Project - ResNet50

Among many Convolutional Neural Networks, Residual Network (ResNet) has been used as the base model for most models that have achieved high recognition accuracy in recent studies. The aim of the project is to train an image classification model for plant diseases with plant leave images using ResNet to prototype a mobile app. The challenge is to develop a practical classifier that can deliver high classification accuracy in recognition of 26 different diseases on the image dataset. The figure below shows the configuration of ResNet50.

---

Method - Plantvillage Dataset

The PlantVillage dataset is used for training the ResNet50. It includes 26 diseases and 12 healthy classes on 14 crops. Dataset images for each disease class are bulked up by Image Rotation method. 1,000 images for each class are randomly selected, then these are divided into 800 and 200 images for the Training and the Testing, respectively.

---

Overall Accuracy

An overall accuracy is 99.5%, and a 100% accuracy is achieved for 22 classes.

---

Implemented on Mobile Application

The trained model with a 99.5% accuracy has been implemented on iPhone 7 Plus using MATLAB Mobile .

---

Demonstration Movie

Here is the demonstration with the trained model implemented as a mobile app on a smarphone.

---

See How You Create Your Own Model

If you want to make your own detector, see and check MATLAB code at my Github repository.