Estimativa de nitrogênio em Annona emarginata (Schltdl.) H. Rainer utilizando espectroscopia no infravermelho próximo (NIRS): Uma abordagem estatística e computacional
Carregando...
Data
2020-02-18
Autores
Gomes, Rafaela Lanças
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O nitrogênio é um elemento mineral essencial para as plantas. Sua deficiência em fases iniciais do desenvolvimento pode gerar alterações fisiológicas e morfológicas que reduzem o crescimento e conflui na não expressão total do potencial genético vegetal. As técnicas mais difundidas para a quantificação do N nas plantas demandam tempo, são destrutivas e liberam compostos tóxicos para o ambiente. A NIRS (Near-Infrared Spectroscopy - Espectroscopia no Infravermelho Próximo), se apresenta como uma técnica alternativa, sendo indireta, mas instantânea, não destrutiva e que não utiliza reagentes químicos, mas necessita de calibração, que pode ser feita por métodos estatísticos e computacionais. Para mudas que são produzidas em viveiros, como as de Annona emarginata (Schltdl.) H. Rainer, é essencial manter o monitoramento de N, de forma rápida e não danosa, para garantir a qualidade e vigor das mudas. Desta forma, este trabalho visou detectar alterações na caracterização espectral foliar de A. emarginata em função do fornecimento de concentrações de nitrogênio e classificar as mudas em função dos níveis de nitrogênio, com base na caracterização espectral, utilizando algoritmos de aprendizado de máquinas e análise estatística multivariada. As mudas de A. emarginata (240) foram mantidas em sistema de hidroponia, com alterações na concentração de nitrogênio: 0 mg.L-1 de N (T1); 52,5 mg.L-1 de N (T2); 105 mg.L-1 de N (T3) e 210 mg.L-1 de N (T4), com 60 repetições (mudas) para cada tratamento. Após 45 dias nestas soluções, foram coletadas três folhas de cada planta, fotografadas e aferidas suas caracterizações espectrais. Foram testadas diversas combinações de técnicas de redução e de classificação estatísticas e de aprendizado de máquinas e propostos índices espectrais com os comprimentos de onda mais selecionados. Quanto às variáveis fisiológicas, foram analisadas as trocas gasosas, fluorescência e pigmentos foliares. Pela caracterização espectral, percebe-se um afastamento total das curvas de mudas com falta e sem falta de N. Conclui-se que a técnica NIRS pode ser aplicada para a detecção de variações na concentração de N nas folhas de plantas jovens (mudas), antes do aparecimento de sinais visuais e que técnicas estatísticas são melhores para se fazer a calibração do método do que técnicas de aprendizado de máquinas.
Nitrogen is an essential mineral element for plants. Its deficiency in early stages of development can lead to physiological and morphological changes that reduce growth and result in the total non-expression of plant genetic potential. The most widespread techniques for quantifying N in plants are time consuming, destructive and release toxic compounds into the environment. Near-Infrared Spectroscopy (NIRS) is an alternative technique, being indirect, but instantaneous, non-destructive and does not use chemical reagents, but needs calibration, which can be done by statistical and computational methods. For seedlings that are produced in nurseries, such as those of Annona emarginata (Schltdl.) H. Rainer, it is essential to keep N monitoring fast and harmless to ensure seedling quality. Thus, this work aimed to detect changes in the leaf spectral characterization of A. emarginata as a function of nitrogen concentration supply and to classify seedlings as a function of nitrogen levels, based on spectral characterization, using machine learning algorithms and multivariate statistical analysis. 240 A. emarginata seedlings were maintained in a hydroponic system, with modifications in nitrogen concentration: 0 mg.L-1 of N (T1); 52.5 mg.L-1 of N (T2); 105 mg.L-1 N (T3) and 210 mg.L-1 N (T4), with 60 replications (seedlings) for each treatment. After 45 days in these solutions, three leaves of each plant were collected, photographed and their spectral characterizations was measured. Several combinations of reduction and classification techniques and machine learning techniques were tested and proposed spectral indexes. Regarding the physiological variables, gas exchange, fluorescence and leaf pigment were analyzed. Due to the spectral characterization, it is possible to observe a total deviation of the curves with the N lacking and non-lacking seedlings. It can be concluded that the NIRS technique can be applied to detect variations in N concentration in the leaves of seedlings before visual symptoms appear, and which statistical techniques are better for method calibration than machine learning techniques.
Nitrogen is an essential mineral element for plants. Its deficiency in early stages of development can lead to physiological and morphological changes that reduce growth and result in the total non-expression of plant genetic potential. The most widespread techniques for quantifying N in plants are time consuming, destructive and release toxic compounds into the environment. Near-Infrared Spectroscopy (NIRS) is an alternative technique, being indirect, but instantaneous, non-destructive and does not use chemical reagents, but needs calibration, which can be done by statistical and computational methods. For seedlings that are produced in nurseries, such as those of Annona emarginata (Schltdl.) H. Rainer, it is essential to keep N monitoring fast and harmless to ensure seedling quality. Thus, this work aimed to detect changes in the leaf spectral characterization of A. emarginata as a function of nitrogen concentration supply and to classify seedlings as a function of nitrogen levels, based on spectral characterization, using machine learning algorithms and multivariate statistical analysis. 240 A. emarginata seedlings were maintained in a hydroponic system, with modifications in nitrogen concentration: 0 mg.L-1 of N (T1); 52.5 mg.L-1 of N (T2); 105 mg.L-1 N (T3) and 210 mg.L-1 N (T4), with 60 replications (seedlings) for each treatment. After 45 days in these solutions, three leaves of each plant were collected, photographed and their spectral characterizations was measured. Several combinations of reduction and classification techniques and machine learning techniques were tested and proposed spectral indexes. Regarding the physiological variables, gas exchange, fluorescence and leaf pigment were analyzed. Due to the spectral characterization, it is possible to observe a total deviation of the curves with the N lacking and non-lacking seedlings. It can be concluded that the NIRS technique can be applied to detect variations in N concentration in the leaves of seedlings before visual symptoms appear, and which statistical techniques are better for method calibration than machine learning techniques.
Descrição
Palavras-chave
Redes neurais artificias, Aprendizado de máquinas, Perceptron multicamadas, Nutrição mineral de plantas, Bootstrap, Índice espectral de vegetação, Artificial neural netwoerks, Machine learning, Multilayer perceptron