Özet:

Deep learning methods have become an important technology in computer vision and image processing, especially in the problem of image classification. One of the most important reasons for this is the superior success that deep learning on different problems has shown. With the development of the internet, very large datasets are gathered and with high-power graphics processor cards, these data can be processed in real time. However, collecting large-scale data in this way for every problem is quite expensive process. For this purpose, pre-trained deep learning models can be used to classify data in lower sizes with the transfer learning method. In this study, a system was developed that was embedded and capable of working in real-time with the transfer learning method to classify objects containing threat elements in images taken from X-ray devices. This system uses the deep learning model of Alexnet, which is an evolutionary nerve network on the NVIDIA Jetson TX2 development card. With this model, the X-ray baggage safety images classify objects containing threatening elements such as knife, weapon, jewel and the Ninja star.   In the created experiment environment, Alexnet was trained with 12,000 images and tested in a real environment. The performance of the recommended method was compared with different computer vision techniques previously performed on the same data set and more successful results were obtained.

Anahtar Kelimeler: