Analysis of fuzzy images by close-up photography of large industrial drones

– Tea quality prediction using a multi-spectral camera –

抄録

Our research is to recognize the tea leaves opening stage the Deep Learning image analysis. Since the quality of tea depends on the stage of the growth, it is therefore important to predict the leaf opening period. Relative amounts of amino acid and theanine has significant effect on the quality of tea. High quality plucked tea leaves contain the maximum level of theanine. However, over time theanine changes to catechin an astringent ingredient in the sunlight. This means the content of the “Umami” ingredients is reduced. The hypothesis in this study is Umami’s level changes over time can be predicted by image analysis. Image analysis is performed using the Continuous Wavelet Decomposition (CWD), and the Deep Learning (DCGAN, PCA, SAE, and LSTM) as methods. We combine these in certain order and use them in analysis. The advantage of with combine 5 methods grades “fuzzy” tea photo images, difficult to classify accurately, than with one single method, as spectrum analysis, AKAZE and so on. By developing an iPhone application that feed back the analysis predict the optimal picking time, it can contribute to the tea quality prediction of large tea farm a large-scale.