This repository contains projects that use models created from deep learning frameworks in image/object classification and detection.
Digit Recognition : The images for the digit recognition project was obtained from Kaggle, consisting of gray-scale images of hand-drawn digits, from zero through nine. Each image was 784 pixels (28 x 28 in dimension). A simple convoluted neural network was developed using the training set and consequently used to predict the digits in the hand-drawn images of the testing set. The neural network achieved a 98.76% accuracy in predicting the hand-drawn digits in the images of the testing dataset.
Pet Classification : The images for the pet classification project was obtained from Kaggle. The images consisted of colored images two pet types - cats and dogs. The images had varying dimensions and were imported at a target size of 256 X 256. Two convoluted neural networks were developed using the training set and consequently used to predict what pet images were in the testing set. The first neural network was self-contructed while the second was built upon a pre-trained image classification model. The self-constructed neural network had three convoluted and three pooling layers within the hidden layer and achieved an 82% accuracy in predicting the pet images of the testing dataset. On the other hand, the neural network built on the pre-trained model achieved a 99% accuracy in predicting the pet images of the testing dataset. To further test the models, I downloaded 10 random images from X, and the self-constructed model correctly predicted 7 of the 12 pet images correctly while the model built on the pre-trained model correctly predicted 10 of the 12 pet images correctly.
Plant Disease Detection : The images for the pet classification project was obtained from