Satellite-image-based crop identification using unsupervised machine learning techniques: Preliminary results
Abstract:
Artificial vision and image processing have been widely used in the field of scientific research related to satellite landscapes with purposes, like soil classification, detection of changes in urban and rural areas, among others. The existing prototypes have reported meaningful results, notwithstanding, the implementation of a system more properly fitting the nature of the images by taking into account factors such as lighting control, noise reduction and presence of clouds is still an open and of-great-interest problem. This paper presents an initial satellite image processing methodology for clustering crops. The proposed methodology is as follows: Firstly, data pre-processing is carried out, followed by a feature extraction stage. Secondly, image clustering is performed by means of a probabilistic algorithm. This methodology is validated with the Campo Verde database built over crops from a Brazil’s area. Our approach reaches a classification percentage 87.97%, sensitivity 87.1%, specificity 97.22 and f1_score 71.78 %.
Año de publicación:
2019
Keywords:
- Satellite image
- Parzen’s probability density function
- Max-min algorithm
- Landsat satellite
Fuente:
Tipo de documento:
Article
Estado:
Acceso restringido
Áreas de conocimiento:
- Aprendizaje automático
- Ciencias de la computación
Áreas temáticas:
- Ciencias de la computación