Week5 — Plant Disease Detection

Theme: Classification to plants that healthy or diseased and predict to photographed plant disease.

Team Members: Sevda Sayan, Fatmanur Turhan, İsmet Seyhan

Last week, we started to experiment with VGG16 and Resnet50 architecture and the training mechanisms we used, Transfer Learning and Training from scratch. We completed the Vgg16 tests. Our success rate with Vgg16 architecture was not what we wanted.

This week we will do the same experiments with Restnet50 architecture.

Resnet (residual neural network) is an improved version of convolutional neural network (CNN). The Resnet model aims to solve the degradation problem of CNN networks. When deep webs begin to converge, the problem of deterioration arises. As the network depth increases, its accuracy reaches saturation but then tends to decline rapidly. ¨ Resnet adds shortcuts between layers to solve this problem. This simple idea prevents corruption as the network deepens. In addition, the Resnet model uses bottleneck blocks for faster training. Resnet50 is a 50-tier network trained on the ImageNet dataset. ImageNet is an image database of more than 14 million images of more than 20,000 categories created for image recognition competitions. Resnet model uses 2 (3x3) convolution layers instead of using (1x1), (3x3), (1x1)

Configurations About Training mechanism

1. Transfer Learning

2. Training from Scratch

To begin with, we trained Restnet50 from scratch using our minimized data set, which we used to train the vgg16 model. The results made us happy. As we can see in the graphs below, we managed to increase the success range from 4% to 16% in the vgg16 architecture up to 94%.

Then we have trained RestNet50 using transfer learning on our dataset. Likewise, we achieved better results than VGG16. Our success rate reached 98% and we reached a satisfactory result.

As a result of our experiments, we decided to use RESTNET 50 architecture using transfer learning method.

Thank you for reading…