Region-based Convolutional Neural Network as Object Detection in Images
Rema Ibrahim Hamad Abushahma, Musab A. M. Ali, Omar Ismael Al-Sanjary, Nooritawati Md Tahir · 2019
In this study, the Region-based Convolutional Neural Network (R-CNN) is utilised for object detection in the MATLAB simulation environment. This involves the basic R-CNN model that uses the SVM algorithm for classification. Firstly, total images of 911 acquired and captured from desert environment acted as the database in this study. The reason for choosing this type of images are due to the unique features specifically variances between objects and background and this is indeed different from city or urban environment images. Next, from these 911 images, 70% are used as training data for the basic R-CNN model and the remainder as testing. Experimental results showed that the accuracy is indeed promising with 83.1% for single object detection, 77.8% for two objects detection and 67.6% for three objects detection with average accuracy of 69%.