Taxonomic determination and physiological analysis of ornamental plants using convolutional neural networks

Abstract

Phenomics is an interdisciplinary scientific field, the object of research of which are phenotypes, their qualitative and quantitative parameters, as well as regularities of their formation during ontogenesis and as a result of interaction with external factors. The methodology of phenomics is a set of approaches for the phenotyping of plants, including a number of the most modern technologies of imaging, spectral analysis, biochemical, molecular and genetic analyses, and also innovative informatics techniques such as image recognition, computer vision and machine learning. The purpose of this work was to develop a phenomics application based on computer vision and methods of machine learning for taxonomic classification and determination of physiological condition of different ornamental plants. As a result of this work, the annotated databases Thuja occidentalis L., Forsythia intermedia Vahl, Heuchera micrantha Douglas ex Lindl., Syringa vulgaris L., Phalaenopsis × hybridum Blume, etc. were created and annotated. The model of a convolution neural network for taxonomic classification and determination of physiological condition of plants on the basis of RGB-images was developed. The training used images obtained in standardized conditions by high quality RGB-cameras. The neural network showed high efficiency of recognition, when analysing with taxonomic properties of decorative plants (about 90.8 %). The developed neural network also demonstrated coefficient of determination (R2 ) about 0.66 in the analysis of physiological state.

ACCURACY ASSESSMENT OF Syringa vulgaris L. MORPHOLOGICAL SIGNS PHENOTYPING WITH A LASER 3D SCANNER PlantEye F500 DEPENDING ON PLANT LOCATION ON THE SCANNED SURFACE

“…Полученные данные загружаются в компьютер, и существует возможность оценки динамики изменений морфологических параметров во времени. Недостатками следует считать отсутствие протоколов проведения феноскрининга морфологических параметров для разных культур (41). Следует учитывать, что при одновременном изучении нескольких растений на сканируемой поверхности увеличивается вероятность ошибки измерений, на что указывают полученные нами экспериментальные данные.…”