Aplicando Mineração de Imagens na Agricultura de Precisão

Abstract

Crop image analysis are consolidated in the precision farming market. In this sense, the use of image processing techniques, image mining and artificial intelligence are fundamental tools. Being able to apply these techniques individually or together. A common problem in image analysis is that small changes in lighting and timing can influence how computational techniques identify its elements. The cost is too high or even unfeasible to universally identify or segment an image. As such, a solid starting point is needed to guide existing techniques. This study presents an experiment using image mining techniques, associated with custom association algorithms. Using expert knowledge to create and label pixel set of interest. Thus, when processing an image, the classes of interest are easily identified and adjusted for each reality. The empirical results indicate that our solution improves the way of selecting patterns by identifying the classes of interest, correctly identifying soil and vegetation. Tests were performed on seven different mosaics from the same culture. The process of identifying the desired classes (soil, plantation) occurred satisfactorily, thus validating our study as a viable solution for precision agriculture.