Abstract
Antipersonnel landmines represent a very serious hazard endangering the lives of many people living in armed conflict counties. The huge number of human lives lost due to this phenomenon has been a strong motivation for this research. Deep Learning (DL) is considered a very useful tool in object detection, image classification, face recognition and other computer vision activities. This paper focuses on DL for the problem of landmines recognition in order to identify its type based on shape features. This research work consists of several stages: gathering a new dataset of Anti-Personnel Mines (APMs) images for training and testing purposes, employing several augmentation strategies to boost the diversity of training data, applying four different Convolutional Neural Network (CNN) models namely VGG, ResNet, MiniGoogleNet and MobileNet, and evaluating their performances on APMs recognition. In conclusion, results indicate that MiniGoogleNet exceed all of other three models in recognizing APMs with the highest accuracy rate of 97%.