A Hand Gesture Recognition Based on SqueezeNet and Octave Convolution Feature Extraction Approach

In recent days, sign recognition plays an important role in overcoming the communication barriers between hearing-impaired community and the normal society where sign language is essential. This paper proposes an efficient feature extraction and feature mapping technique by employing SqueezeNet fusion Octave Convolution method (SqueezeNet fusion OctConv). Initially, the input image data is preprocessed to transform the data to a common scale and avoid imbalance of model. The SqueezeNet model with fire module is employed for efficient parameter reduction and significant feature extraction. The Octave Convolutional (OctConv) technique is applied to refine the extracted features that optimizes the feature utilization, handles spatial components and reduces redundancy which enhances the feature representation. The integration of fully connected layers with spatial attention mechanism enables the model to learn complex relationships and assign weights based on relevance ensuring precise classification. The proposed method is evaluated on NUS-II dataset and achieved high accuracy of 99.25% which surpassed the existing sign recognition models such as Intensive Feature Extrication Deep Network, Densely Connected Res2Net.