An optimization method of machine learning model for image recognition by transferring images in grayscale

When applying image recognition based on machine learning, TensorFlow framework based on convolutional neural network is often taken in consideration. Generally speaking, current image recognition program designed on TensorFlow are comprehensive enough with high accuracy in identifying images. However, when analyzing blurry images with less targets to recognize, existing models may not work as accurately as expected. This study aims at providing ideas for optimizing machine learning model for image recognition by transferring images in grayscale. In this article, animal data set was applied as test data to analyze since animal images are usually simple and concise which exactly meet our requirements. The study compared the prediction outcomes by the scale of accuracy. As the test showed, the proposed model archived better results using gray scaled image for classifying blurred image with the accuracy of the prediction raised from 33.03% to 94.01% in one test. The result suggested that applying grayscale images to improve the accuracy of blurry image recognition based on machine learning is a feasible and efficient proposal.