Analysis of fuzzy tea leaf images to predict a quality by DCGAN

Purpose of our research is to classify tea image at each shooting time.The flavor and quality of tea leaves change greatly depending on the time of picking.The more Umami ingredient “theanine”, the higher the quality.The more Astringent ingredient “catechin”, the lower the quality.Therefore, it is important to pick the tea at the timing when the theanine content is maximized.In an actual farm, the timing of picking is decided at the leaf opening stage, not at the component analysis.The leaf opening period consists of 8 stages, from the germination period until the 7th leafopening period.As a rule of thumb, the highest quality is achieved at the 5th leaf opening stage.Skilled technique is required to determine the leaf stage.We shoot a tea plantation with a large drone and analyze the shot image by deep learning.Analyzing each leaf opening stage leads to quality prediction.And I've been thinking about how to predict quality.In order to determine the leaf opening stage, we tried to classify the images using a well-known method, Luminance histogram,Spectrum analysis by FFT and AKAZE.However, none of the conventional methods were able to classify well.The cause is that it is difficult to extract the features due to the fuzzy of the tea image.Therefore, we devised a classification method using deep learning.We use DCGAN, SAE, LSTM and CWD.The originality of this research lies in the fact that these analyzes are not performed individually but continuously.