Autism is a sneaky developmental condition characterised by delayed communication and social interaction improvement. The number of autistic children and adults is growing by the day. Because the causes of autism are unknown, early detection and intense treatment can make a significant difference. This illness causes significant behavioural changes in children and adults. With the advent of artificial intelligence, this is now achievable, potentially saving the lives of many individuals. The use of transfer learning to detect ASD in youngsters is proposed in this study. The suggested methodology detects autism using seven alternative CNN architectures: Resnet50 model , the MobileNet, Inception , VGG16, ResNet50v2 , Xception , and NasNetmobile model models . The Adam optimization methodology is a simple process that outperforms the existing method in terms of performance. As training and testing data, a dataset containing images of structural magnetic resonance imaging (sMRI) for children with autism and non-autism is provided. In comparison to seven models of architecture gave an accuracy of 0.992 for the MobileNet model.