Scholarlink

<p>Autism spectrum disorder (ASD) is a complex psychological disease described by persistent problems in restricted behaviours, speech, social interaction, and non-verbal communication. Individuals with ASDs have trouble identifying and connecting with others. The ASD symptoms might arise in an extensive range of conditions. Many dissimilar kinds of functions exist for individuals with ASD. Emerging expert methods for recognising ASD that depend on the facial signs of children are one of the foremost contributions for perceiving ASD at an initial phase. At present, deep learning (DL) models have delivered excellent performance in a diversity of pattern recognition tasks. The utilization of models that depend on convolutional neural networks (CNNs) was projected by numerous researchers to be utilized in the study of ASD. In this paper, an innovative methodology called Deep Feature Fusion Integrated with Wavelet Neural Network for Autism Spectrum Disorder Identification (DFFWNN-ASDI) utilizing facial images is proposed. The primary goal of the DFFWNN-ASDI technique is to serve as a promising diagnostic aid in clinical settings for the early and non-invasive detection of autism. Initially, the DFFWNN-ASDI technique performs pre-processing by utilising the Gaussian Gabor Filter (GGF) to enhance relevant facial features and suppress noise. Furthermore, a feature fusion strategy is implemented by incorporating deep features extracted from VGG16, DenseNet-121, and XceptionNet models, ensuring a rich and diverse representation of facial cues relevant to ASD. Moreover, the wavelet neural network (WNN) technique is employed for detecting ASD. Additionally, the chimp optimisation algorithm (COA) technique is implemented for optimal tuning of WNN parameters to enhance classification performance. Eventually, the Grad-CAM + + technique is applied to provide visual interpretation and abnormality detection. To establish the improved performance of the DFFWNN-ASDI approach, an extensive set of simulations was conducted under the autism image dataset. The comparison outcomes of the DFFWNN-ASDI model established a greater accuracy value of 98.42% compared to existing approaches across diverse metrics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multimodal fusion of deep transfer learning with wavelet neural network for autism spectrum disorder detection using facial images

  • 期刊论文

  • 发表日期：

    2026年3月20日

查看全文
  • Jahanara Shaik,
  • R. Shekhar,
  • Chetan J. Shelke

摘要

Autism spectrum disorder (ASD) is a complex psychological disease described by persistent problems in restricted behaviours, speech, social interaction, and non-verbal communication. Individuals with ASDs have trouble identifying and connecting with others. The ASD symptoms might arise in an extensive range of conditions. Many dissimilar kinds of functions exist for individuals with ASD. Emerging expert methods for recognising ASD that depend on the facial signs of children are one of the foremost contributions for perceiving ASD at an initial phase. At present, deep learning (DL) models have delivered excellent performance in a diversity of pattern recognition tasks. The utilization of models that depend on convolutional neural networks (CNNs) was projected by numerous researchers to be utilized in the study of ASD. In this paper, an innovative methodology called Deep Feature Fusion Integrated with Wavelet Neural Network for Autism Spectrum Disorder Identification (DFFWNN-ASDI) utilizing facial images is proposed. The primary goal of the DFFWNN-ASDI technique is to serve as a promising diagnostic aid in clinical settings for the early and non-invasive detection of autism. Initially, the DFFWNN-ASDI technique performs pre-processing by utilising the Gaussian Gabor Filter (GGF) to enhance relevant facial features and suppress noise. Furthermore, a feature fusion strategy is implemented by incorporating deep features extracted from VGG16, DenseNet-121, and XceptionNet models, ensuring a rich and diverse representation of facial cues relevant to ASD. Moreover, the wavelet neural network (WNN) technique is employed for detecting ASD. Additionally, the chimp optimisation algorithm (COA) technique is implemented for optimal tuning of WNN parameters to enhance classification performance. Eventually, the Grad-CAM + + technique is applied to provide visual interpretation and abnormality detection. To establish the improved performance of the DFFWNN-ASDI approach, an extensive set of simulations was conducted under the autism image dataset. The comparison outcomes of the DFFWNN-ASDI model established a greater accuracy value of 98.42% compared to existing approaches across diverse metrics.