| Sumario: | Autism is an insidious developmental disorder exemplified by impaired development in communication and social interaction. The number of cases with autism in children and adults are increasing day by day. The causes of autism are unknown, hence the early diagnosis of autism accompanied with intensive treatment can make a wide behavioral change in the lives of children or adults with this disorder. With the advent of artificial intelligence this has become possible thus saving lives of many people. This paper proposes the detection of ASD in children with the help of transfer learning. The proposed methodology uses four different CNN architecture in the detection of autism namely, VGG19, Resnet50, InceptionV3 and NASNetLarge models. A dataset consisting of images of the facial expressions of children with autism and non-autism are provided as training, testing and validation data. The architecture NASNetLarge provided an accuracy of 87.50% and a loss of 0.372 compared to other three models.
|
|---|