Processamento de imagens hiperespectrais para diferenciação dos agentes causais da antracnose e ramulose do algodoeiro.

Abstract

This work describes a new method for identification and classification as the precise and rapid C. gossypii (CG) and C. gossypii var. cephalosporioides (CGC) grown in culture medium using hyperspectral NIR images and multivariate pattern recognition. Five isolates were employees of CG and 46 of CGC. The different isolates of CG and CGC were cultured in Czapek agar with 12 h photoperiod for 15 days. The spectral measurements were performed between 1000 and 2500nm, with 256 bands with a highperformance camera with a 50 mm lens. The pattern recognition models were developed APS LDA and SIMCA strategies. The samples were separated using the sample selection algorithm in three sets the number of samples: 3, 1 and 1 for CG and 20, 8, 18 to CGC, totaling 23 (Test), 9 (Validation) and 19 (PREDICTION ) totaling 51 samples of CG and CGC fungi. The prediction results showed a type II error for the CG set in the SIMCA model. As with the APS-LDA there were misclassifications. CGC conjunction with a number of 18 prediction samples were not observed errors of type I and II using the strategies APS-LDA and SIMCA. A parallel validation of the results obtained with APS-LDA was performed with BoxPlot analysis with the variables selected by the APS. The results were consistent with no evidence of outlies. Therefore, a new procedure HSI and APS-LDA for identification and classification of etiliológicos agents CG and CGG is proposed with the advantages of greater analytical capacity with multiple analyzes, lower cost observing large numbers of samples, speed and ease of implementation.