Enhancing Ann-Euclide distance hybrid algorithm for object recognition: Empowering individuals with visual impairment

ABSTRACT: Machine learning technology provides many exceptional tools for making the lives of visually impaired individuals much more accessible and enabling them to interact with their respective environments. The hybrid algorithm, which consists of Artificial Neural Networks (ANN) and Euclidean Distance, is used in an object recognition application. In line with this, the study focuses on enhancing the Artificial Neural Network (ANN) and Euclidean Distance Hybrid Algorithm to address feature extraction failure, scalability, and overfitting issues. The object recognition utilizes the ANN and Euclidean Distance Hybrid Algorithm, wherein the ANN is used for training and detection, while the Euclidean Distance calculates the error between values. The enhanced hybrid algorithm incorporates complex pre-processing strategies, including noise reduction, iterative normalization, and the Sliding-Window Weight Fusion (SWWF) method. Moreover, it also utilized DeepMultibox to enable the system to prepare for multiple bounding boxes within a single image and Gradient Augmentation to improve the generalization capability of the ANN-Euclidean Distance Hybrid Algorithm. To further assess the effectiveness of the enhancement, both algorithms applied a splitting technique to split the dataset for training and testing. Moreover, the two algorithms test their performance, including accuracy, precision, recall, F1 Score, and computation time with the same data set, which are CIF AR-10 and CIFAR-100. The result indicates improvement due to feature extraction, enhanced accuracy, reduced computational time, and better generalization of the hybrid algorithm through the integration of the enhancements, wherein it reduces the misclassifications, making the Enhanced ANN and Euclidean Distance Hybrid Algorithm reliable for real-time application.