Binary Classification of Criminal Tools from the Images of the Case Using CNN
Mustafa Kaya, Betül Ay, Serkan Karakuş · 2018
Recent advances in object detection and classification has accelerated by virtue of Convolutional Neural Networks (CNN) and large-scale image datasets like ImageNet. In this study, we aim to classify objects such as knives and weapons on the image data contained in electronic evidence, which may be criminal objects in judicial cases. To this end, we make this process autonomous by contributing to the examination of electronic evidence via artificial intelligence. For training of the CNN model presented in this paper, we use 10000 training data handled from ImageNet. The model with best parameter settings has been achieved 94.37% classification accuracy on 2500 test samples. We also investigate the effect of different parameters used in the network to increase the classification accuracy and try to achieve a set of optimal hyper parameters by tuning best possible parameters. As a result of the work, we present a detailed comparative study on binary object classification.