LAUSR

The use of deep learning using image data has already made a lot of progress. The main function is to track or classify related objects with the model of the data set in which the learning has been conducted in the image and it is used in various fields. Therefore, the accuracy has improved a lot, but the proportion of hardware use increases rapidly because the depth of the model is increased and utilized. In this paper, we propose a process to improve results by using the lightweight model and additional procedures in the program in systems using deep learning. In the proposed system, an additional image processing process is performed to increase accuracy in the results after deep learning processing, unlike other systems. It uses residual CPU usage in desktop or embedded environments with GPU, maintains processing speed, and introduces additional image processing processes to mitigate graphic memory usage using lightweight deep learning models. As a result, a system that improves the accuracy of the object area by 6% based on red apples and 16% based on green apples can be designed to reduce hardware costs and increase performance.