Modelo de aprendizaje no supervisado para el cálculo de atributos estructurales de cultivos de papa empleando técnicas de fotogrametría.
Fecha
Autor corporativo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Compartir
Director
Altmetric
Resumen
The main objective of this work is to train and test a model based on unsupervised learning that allows segmenting the areas of interest of the data, in order to mathematically calculate the structural attributes of the plant such as surface and height. Two machine learning methods, nine image features including position (x, y and z), colors (R,G and B) and normal parameters were tested for segmentation of soil and individual plants. K-means clustering obtained the highest percentages of correctness (86.33%) for the separation of plants and soil. The developed method was validated by a case study in which the dataset constructed by GITUD previously sampled a crop of semi-mature plants. Therefore, individual plants were automatically separated and the characteristics of each plant (height and shoot area) were successfully extracted using the proposed automatic segmentation line. Finally, the limitation of this study is that the proposed methods have been developed and evaluated for potato in vegetative state. However, the application of segmentation to evaluate the vegetation state of a different plant is open for model retraining. Over time, the accuracy of segmentation will be improved with other clustering methods and how structural attributes are calculated through experimentation or further iterations to obtain a more accurate measure of the shoot area attribute. Vegetation condition plays a key role in determining crop condition and allows action to be taken in case of abnormal behavior in the development of cultivated plants. In crops of more than one hectare, determining the state of the vegetation is a challenge due to its large size, monitoring currently involves specialized personnel in the field and high costs in the collection and analysis of crop samples, in addition to entering areas of difficult access such as those of the Colombian terrain, so there is a need for other large-scale data collection techniques such as: the study of satellite images or point clouds taken with unmanned aerial vehicles (UAV).