Self-Learning Model for Pattern Recognition in Vision System Based on Adaptive Kernel

Recently, the solution for recognizing and understanding an object based on visuals is to integrate the adaptation function (continuous machine-driven process) into the system update function involving humans (continuous human-driven process). However, this has created a gap between the adaptation function and the system. This situation requires understanding the system viewed as a dynamic composition of the learning process. This research introduced a self-learning model in the form of an adaptive kernel equipped with the SpinalNet architecture, and the goal of this study is to increase the Convolutional Neural Network (CNN) accuracy. The model consisted of a domain model, contextual knowledge, and adaptive learner developed based on the CNN with SpinalNet. The combination of Adaptive Kernel and SpiralNet in this CNN has a significant impact, allowing the model to adjust the selection of subsequent kernels based on the optimal input from the previous kernel. Moreover, this combination results in lower memory usage during training. The evaluation results show that our proposed model provides better classification accuracy than the SpiralNet model without the Adaptive Kernel. Furthermore, in terms of inference speed, our model outperforms SpiralNet, as evidenced by the use of fewer parameters