Reinforced Deep Learning By Discriminant Feature Trace Transform

Classification of objects using YOLO deep learning is currently very popular. When the discovered object is rotated, however, the pre-trained YOLO model frequently generates incorrect answer predictions. In order to improve the accuracy of the YOLO deep learning pre-trained model, we present in this paper a method for rotating an object's image back to its initial gain using the DFTF algorithm. The DFTF algorithm is designed to take the results of the trace transform and accumulate the values for each column in order to predict the object's degree of rotation. The results of this calculation are then used to rotate the object back to its normal axis. The experiment is therefore a collaboration between YOLO and DFTF with the aim of more accurately predicting the object's responses. Five groups of data will be chosen for experimentation, each of which can be used to determine the precision of the experimental image with rotational issues. The average of accuracy is 98.56%.