Diagnosis of nitrogen levels in bean leaves using computer vision and artificial neural networks

Abstract

Determination of nitrogen (N) levels in beans culture are slow or depend on an experienced professional, so the objective of this work is to aid in diagnosis nitrogen levels through of Computational Vision and Artificial Neural Networks (RNA). Beans were grown over 5 doses of N (50, 100, 150, 200 and 250 mg L-1). The data of chlorophyll and N contents and leaves images was realized 45 and 58 days after sowing. To make the diagnosis were used Gray Level Co-Occurrence Matrix (GLCM), non-texturing Statistic and Local Binary Pattern (LBP) for the training and test the Artificial Neural Networks (ANN) and Multilayer Perception for regression and later classification of the levels of N. This work demonstrated that the three methods are promising for determining the levels of N in bean leaves with correlation coefficients greater than 0.7, with GLCM having the best correlation coefficient, 0.74. The combination of the 3 methods provides even better results, with the coefficient rising to 0.76. If the purpose is to diagnose the levels of N in classes, which would help to define whether the levels are adequate or not, there is a percentage of correctness of 81.12%.