الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 86 |  |  |
| | | from | 86 | | |
Abstract
Classifying objects is a complex problem in the field of computer vision. The Deep Learning (DL) algorithm is a computerized model that simulates the human brain functions and operations. However, training the DL model is a costly process in machine resources and time. Therefore, investigating the performance of the DL algorithm is mainly needed. Thesis investigates the performance of the DL algorithm for two case studies and an application work. The first case study is multi-class Image classification using CNN. The application work is using CNN to classify agricultural products. The second case study is evaluation of YOLOv3 algorithm for object Detection and classification.
The convolutional neural network (CNN) is mostly used to build a structure of the DL models. In the first experiment the CNN model pre-trained on the Image-Net data set is used to classify images of the PASCAL VOC 2007 data set. The transfer learning approach is used to improve the performance of the CNN model, where classification works reasonably well with less computation time and fewer machine resources. The behaviour of the model is studied, and the performance has been measured. The obtained results are compared with the obtained test results from the Super-vector coding of local image descriptors method, SVM method, and Region Ranking SVM method, The final results evaluate the DL algorithm.
Apply the DL algorithm to classification Egyptian vegetables and fruit is mainly needed. The application work is limited to apply a classification based on the DL algorithm with four classes of Egyptian vegetables and fruits. The CNN model has been trained using the Amazon web service (AWS). The data set for the training and test process is collected from the Egyptian market. The system’s value increased as much disease infection happens due to human interaction with vegetables and fruits.
You Only Look Once version 3 (YOLOv3) is a DL model for object detection and classification. It is a single neural network architecture model that simultaneously uses features of images and simultaneously predicts the bounding box for all image classes. In the second experiment, the DL model based on YOLOv3 architecture is implemented using Tensor-Flow as a framework. The training process had been done using the PASCAL VOC 2007 data set and the PASCAL VOC 2012 data set, using The Adaptive Moment Estimation Optimizer (Adam optimizer). The trained model is tested using the PASCAL VOC 2007 test data set. The final results, evaluate the YOLOv3 model performance for object detection and classification.