Home // International Journal On Advances in Systems and Measurements, volume 17, numbers 3 and 4, 2024 // View article
Authors:
Alex Bertolla
Paulo Cruvinel
Keywords: ccd sensor; digital image; feature extraction; dimensionality reduction; principal component analysis
Abstract:
The pest control in agriculture based on digital imaging sensors has increased significantly in the past decades. Such a strategy has become possible due to the continuous improvements in computational intelligence and machine learning techniques. However, the demand for analyzing and processing such an amount of data generated by these sensors has become a challenge due to the high dimensionality. This article presents a study on the dimensionality reduction of features from digital images acquired with a Charge-Coupled Devices sensor in an agricultural field, to choose the optimal number of principal components for reducing feature dimensionality. It also presents a machine-learning method for the pattern recognition of this species of caterpillar (Fall armyworms - Spodoptera frugiperda) in its different growth stages. In such a context, selecting the optimal number of principal components for dimensionality reduction, retaining only the necessary information associated with the main variables that describe the object of interest. The results have shown that using Hu invariant moments for feature extraction, dimensionality reduction was possible for all analyzed cases, leading to 80% of the original data. In this context, it was possible to preserve the semantic characteristics collected by the sensor. Support Vector Machine classifiers have reached more than 70% of accuracy and more than 80% of precision. Moreover, the performance of the classifiers was 30% faster when working with the dimensionality reduced of the feature vector than when working with the original data.
Pages: 138 to 145
Copyright: Copyright (c) to authors, 2024. Used with permission.
Publication date: December 30, 2024
Published in: journal
ISSN: 1942-261x