AUTOMATIC CAUSE DETECTION  OF HUMAN DEATH USING  CONVOLUTIONAL NEURAL NETWORK

MEBTU, KEFALE ALACHEW

dc.contributor.author	MEBTU, KEFALE ALACHEW
dc.date.accessioned	2022-03-18T06:32:09Z
dc.date.available	2022-03-18T06:32:09Z
dc.date.issued	2021-06
dc.identifier.uri	http://ir.bdu.edu.et/handle/123456789/13209
dc.description.abstract	Nowadays,human violent deaths, suspicious deaths, and sudden deaths occur around the world in both developed and developing countries. Car accident, Suicide death, homicide death, comes first in the list of the loss of one’s life in unexpected human deaths. Forensic pathologists use a laboratory to detect the cause of death. Therefore, we proposed and a developed convolutional neural network model, which is capable of detecting the cause of death in line with forensic pathology science. The proposed system has three components: preprocessing, feature extraction and classification. In image preprocessing, we normalize the image to a needed /standard size. For feature extraction, we use a sequence of convolutional neural networks such as pooling, convolution, activation, dropout, and fully connected layer. Finally, a 3-way Softmax is used for classifying a specific case (death scene investigation case) into a predefined specific class (Car Accident, Homicide, and Suicide)The proposed system is implemented using Keras on Google colab tested using sample image dataset collected from Federal Police and Dagmawi Minillik Referral Hospital about 1200 images in each class. The model achieved a diagnosis accuracy of 99% for training and 98.9% to detect cause of death. Our model was faster to train and had smaller number of parameters as compared to the state-of-the-art models. Besides, it improved the performance of state-of-the-art models as well by 28% (AlexNet) and 18% (GoogLeNet) because of their parameters. Keywords: CNN, Feature Learning, Forensic, Softmax	en_US
dc.language.iso	en_US	en_US
dc.subject	Software Engineering	en_US
dc.title	AUTOMATIC CAUSE DETECTION OF HUMAN DEATH USING CONVOLUTIONAL NEURAL NETWORK	en_US
dc.type	Thesis	en_US