Method of Anomalies Detection in Persea Americana Leaves with Thermal and NGRDI Imagery

Abstract

Among the main agricultural export products of Perú is the Persea Americana (Hass avocado). However, pests and deficiencies cause a decrease in avocado crop yield. For that reason, anomalous visual characteristics detection of a leaf has been studied for thermal and multispectral images. In this study, we propose a procedure for imagery acquisition and evaluate factors that can affect the acquisition of thermal imagery; furthermore, we propose a method for the extraction of characteristics to classify between a healthy leaf and a leaf with some visible anomalies that indicate some deficiency or pest. This method consists of the following stages: segmentation of thermal and NGRDI images, resampling, clustering with k-means algorithm and classification with SVM. Likewise, we test three cases of image acquisition to analyze the effects on statistical descriptors. Finally, the classification stage was tested with leaves processed; as a result, we got out an average accuracy of 82.67%, from ten experiments with the same set of images.