{"id":15742,"date":"2022-06-24T13:44:55","date_gmt":"2022-06-24T06:44:55","guid":{"rendered":"https:\/\/bgrssb.icgbio.ru\/2022\/?p=15742"},"modified":"2022-09-20T10:32:31","modified_gmt":"2022-09-20T03:32:31","slug":"the-system-of-computer-vision-for-extracting-quantitative-characteristics-of-wheat-shoots","status":"publish","type":"post","link":"https:\/\/bgrssb.icgbio.ru\/2022\/2022\/06\/24\/the-system-of-computer-vision-for-extracting-quantitative-characteristics-of-wheat-shoots\/","title":{"rendered":"The system of computer vision for extracting quantitative characteristics of wheat shoots"},"content":{"rendered":"

by Busov Igor | student of NSU<\/em><\/p>\n

Methods of analyzing quantitative characteristics of wheat shoots based on visual or tactile
\nexpert assessment have some disadvantages: subjectivity, labor input. An alternative
\nmethod was implemented in the work, which is based on the analysis of images by computer
\nvision methods.
\nThe following methods and algorithms were used in the work: a convolutional neural
\nnetwork of the Unet architecture, a modified tgi index, with a further search for the
\nmaximum of the objective function using a genetic algorithm, depth-first search (DFS),
\nDijkstra’s algorithm, Student’s criterion, Bartlett’s criterion, K\u00b2-test D ‘Agostino, MannWhitney test, Kruskal-Wallis test, logistic regression, RandomForest machine learning
\nalgorithm, ResNet architecture convolutional neural network with 18, 50 and 101 layers.
\nThe system created in this work was tested on the extraction of two quantitative
\ncharacteristics: ploidy and 1102\/L-25 genotype.
\nFor the task of image segmentation provided for determining ploidy, the results of the best
\nmodels on test samples are 0.8463870901428718 (IoU), for determining the genotype –
\n0.8999361294443262 (IoU). Statistically significant differences in the distributions of
\ndescriptors extracted from images of shoots of different ploidy were almost never found, but
\nstatistically significant differences were found almost everywhere in the distributions of
\ndescriptors of shoots images of different genotypes. The results of the best classification
\nmodels on test samples for solving the problem of determining ploidy were as follows:
\nwithout division into time points – 0.6538461538461539 (accuracy), the first time point –
\n0.6153846153846154 (accuracy), the second time point – 0.7692307692307693 (accuracy),
\nthe difference – 0.46153846153846156 (accuracy). The best genotype classification model,
\n1102\/L-25, predicted the genotype without error.<\/p>\n

\n