Classification and Segmentation Using a Convolutional Neural Network on the Oxford-17 Dataset Written in Matlab
Designed and implemented two convolutional neural networks, using the deep learning toolbox in matlab. The data used was the famous Oxford-17 flower dataset with 17 different species of flowers and 80 images per class. Using my own CNN developed from scratch, I trained one classification model that achieved a 40% mean classification accuracy (classnet.mat) and a segmentationmodel that achieved a mean accuracy of 87.26% and a weighted IoU score of 77.72%. The segmentation model was only based on the daffodil class.
The code for the classification mode is in classification.m and the code for the segmentation model is in segmentation.m. The dataset for the classification CNN is 17flowers and the dataset for segmentation is daffodilSeg.
The other two folders contain output and evaluation images for the segmentation model.
You can load the trained classification or segmentation model (.mat files) to test the networks.