Projeto e caracterização de discriminação de cores utilizando fotodiodo PIN lateral com porta através de aprendizagem de máquina
Carregando...
Arquivos
Citações na Scopus
Tipo de produção
Dissertação
Data
2021
Autores
Silva Júnior, João Batista da
Orientador
Giacomini, R.
Periódico
Título da Revista
ISSN da Revista
Título de Volume
Citação
SILVA JÚNIOR, João Batista da. Projeto e caracterização de discriminação de cores utilizando fotodiodo PIN lateral com porta através de aprendizagem de máquina. 2021. 132 p. Dissertação (Doutorado em Engenharia Elétrica) - Centro Universitário FEI, São Bernardo do Campo, 2021 Disponível em: https://doi.org/10.31414/EE.2021.T.131380.
Texto completo (DOI)
Palavras-chave
fotodiodo PIN,polarização de ponta VG 3,aprendizado de máquina
Resumo
O avanço tecnológico no mercado de sensores de imagem vem crescendo de forma ao longo dos anos com buscas incessantes na detecção da radiação eletromagnética, nas áreas de segurança, medicina, aeroespacial, câmeras fotográficas, televisões, celulares etc. Essa busca motivou a comunidade científica a realizar estudos dos fotodiodos PIN para detecção da radiação eletromagnética, que são dispositivos capazes de converter a energia luminosa em um sinal elétrico. Ainda existem grandes lacunas nos estudos dos fotodetectores, motivando pesquisadores do mundo inteiro a estudá-los. Além do mercado óptico, outra tecnologia imersiva está em ascensão, a IA (Inteligência Artificial), dividida em três grandes grupos: Aprendizado de Máquina (AM), Ciência de Dados e “Big Data”, que permitirá entender e ensinar as tomadas de decisões das máquinas. No mercado de aquisição de imagens, as câmeras de hoje possuem boa resolução, porém, seu modelo de tratativa da informação ainda é limitado, devido à utilização do modelo RGB, que combinas as radiações R (RED – Vermelho) G (GREEN – Verde) e B (BLUE – Azul) para gerar as demais radiações do espectro eletromagnético visível. A captura dessas radiações ocorre através da detecção dos comprimentos de ondas via filtros ópticos discretos depositados na matriz dos fotodetectores. Em virtude das lacunas apresentadas nos estudos dos fotodetectores e da ascensão da IA, este trabalho visa estudar o fotodiodo PIN lateral com porta para discriminar as cores do espectro visível com auxílio dos algoritmos do aprendizado de máquina, no intuito de expandir a qualidade das imagens capitadas. As análises foram divididas em três partes sendo que na primeira parte, o estudo propõe índices de mérito capazes de discriminar as cores através da polarização de porta VG. Os índices de mérito XRGB, YRGB, I’CAT(VG) e ICAT,int demostram essa possibilidade. Para os índices XRGB e YRGB os estudos mostram que a faixa de VG se estende de -0,8V à +0,8V. Já para o método derivativo (I’CAT(VG)) a faixa de VG para discriminar está compreendida entre -0,8V à -0,5V, tendo o ponto máximo de discriminação para VG = -0,65V. Finalizando os estudos dos índices, o método integrativo (ICAT,int(VG)) informa que a melhor faixa para se trabalhar é 0V à 0,5V, onde o comportamento é linear, apresentando como discriminador os coeficientes angulares (BX) de cada comprimento de onda RGB. Na segunda parte, é realizado o estudo dos algoritmos gerados para os dados caracterizados do fotodiodo PIN lateral com porta, sendo: Linear Discriminant, Quadratic Discriminant, Linear SVM, Quadratic SVM, Cubic SVM, Cosine KNN e Ensemble Subspace Discriminant. Todos os modelos durante a fase de treinamento mostraram acurácia acima de 90% e ROC/AUC = 1. Na fase de teste, todos os modelos ficaram com acurácia acima de 94%, sendo que, três deles,
4
Linear Discriminant, Quadratic Discriminant e Quadratic SVM obtiveram 100% de acurácia. Na terceira parte, é proposta a discriminação através da modificação do um circuito eletrônico discutido no item 3.6 deste trabalho, através das polarizações VG que são quatro no total, -1,0 V, -0,6 V, +0,6 V e +1,0 V. Esse circuito fornece a relação entre a tensão de saída V? pelo comprimento de onda ?, através da razão logarítmica entre as correntes dos fotodiodos laterais com porta. A melhor resposta de detecção foi apresentada para VG1 = +1,0 V e VG2 = -0,6 V, para as três potências ópticas em análises (0,1; 0,5 e 1,0 W/m2). As demais combinações de VG demostram resultados satisfatórios para qualquer valor de potência óptica para a faixa de ? entre 200 nm a 500nm, onde existe uma boa variação na tensão de saída (V?) do circuito. O terminal de porta proposto neste dispositivo auxilia positivamente na discriminação dos comprimentos de ondas, abrindo novos horizontes na detecção da radiação eletromagnética
Technological advancement in the imaging market in general has been growing exorbitantly over the years with relentless searches in the detection of electromagnetic radiation, in the areas of security, medicine, aerospace, cameras, televisions, cell phones and etc. This search motivated the scientific community to carry out studies of PIN photodiodes for detecting electromagnetic radiation, which are devices capable of converting light energy into an electrical signal. There are still great gaps in the studies of photodetectors, greatly motivating researchers from all over the world to study them. In addition to the optical market, another immersive technology is on the rise, the AI (Artificial Intelligence), divided into three major groups: Machine Learning (AM), Data Science and “Big Data”, which will allow understanding and teaching of machine decisions. In the image acquisition market, today's cameras have good resolution, however, their model for dealing with information is still limited, due to the use of the RGB model, which combines the primary radiation R (RED - Red) G (GREEN - Green) and B (BLUE - Blue) to generate the other radiations of the visible electromagnetic spectrum. The acquisition of these colors occurs through the detection of wavelengths via discrete optical filters deposited in the photodetector matrix. Due to the gaps presented in the studies of photodetectors and the rise of AI, this work aims to study the lateral PIN gatede photodiode to discriminate the colors of the visible spectrum with the help of machine learning algorithms, in order to expand the quality of the capitulated images. The analyzes were divided into three parts and the first part, the study proposes indications of merits capable of discriminating colors through the gate polarization (VG). The merit indexes XRGB, YRGB, I'CAT (VG) and ICAT, int demonstrate this possibility. For the XRGB and YRGB indices, studies show that the VG range extends from -0.8V to + 0.8V. For the derivative method (I’CAT (VG)), the range of VG to discriminate is between -0.8V to -0.5V, with the maximum point of discrimination for VG = -0.65V. Finishing the studies of the indexes, the integrative method (ICAT, int (VG)) informs that the best range to work on is 0V to 0.5V, where the behavior is linear, presenting the angular coefficients (BX) of each length as discriminator RGB waveform. In the second part, the study of the algorithms generated for the characterized data of the lateral PIN gated photodiode is carried out, being: Linear Discriminant, Quadratic Discriminant, Linear SVM, Quadratic SVM, Cubic SVM, Cosine KNN and Ensemble Subspace Discriminant. All models during the training phase showed accuracy above 90% and ROC / AUC = 1. In the test phase, all models were accurate above 94%, with three of them, Linear Discriminant, Quadratic Discriminant and Quadratic SVM obtained 100% accuracy. In the third part, discrimination is proposed through 6 the modification of an electronic circuit discussed in item 3.6 of this work, through the VG polarizations which are four in total, -1.0 V, -0.6 V, +0.6 V and +1.0 V. This circuit provides the relationship between the output voltage V? by the wavelength ?, through the logarithmic ratio between the currents of the lateral photodiodes with port. The best detection response was presented for VG1 = +1.0 V and VG2 = -0.6 V, for the three optical powers in analysis (0.1; 0.5 and 1.0 W / m2). The other VG combinations show satisfactory results for any value of optical power for the ? range between 200 nm to 500nm, where there is a good variation in the output voltage (V?) of the circuit. The gate terminal proposed in this device positively assists in the discrimination of wavelengths, opening new horizons in the detection of electromagnetic radiation
Technological advancement in the imaging market in general has been growing exorbitantly over the years with relentless searches in the detection of electromagnetic radiation, in the areas of security, medicine, aerospace, cameras, televisions, cell phones and etc. This search motivated the scientific community to carry out studies of PIN photodiodes for detecting electromagnetic radiation, which are devices capable of converting light energy into an electrical signal. There are still great gaps in the studies of photodetectors, greatly motivating researchers from all over the world to study them. In addition to the optical market, another immersive technology is on the rise, the AI (Artificial Intelligence), divided into three major groups: Machine Learning (AM), Data Science and “Big Data”, which will allow understanding and teaching of machine decisions. In the image acquisition market, today's cameras have good resolution, however, their model for dealing with information is still limited, due to the use of the RGB model, which combines the primary radiation R (RED - Red) G (GREEN - Green) and B (BLUE - Blue) to generate the other radiations of the visible electromagnetic spectrum. The acquisition of these colors occurs through the detection of wavelengths via discrete optical filters deposited in the photodetector matrix. Due to the gaps presented in the studies of photodetectors and the rise of AI, this work aims to study the lateral PIN gatede photodiode to discriminate the colors of the visible spectrum with the help of machine learning algorithms, in order to expand the quality of the capitulated images. The analyzes were divided into three parts and the first part, the study proposes indications of merits capable of discriminating colors through the gate polarization (VG). The merit indexes XRGB, YRGB, I'CAT (VG) and ICAT, int demonstrate this possibility. For the XRGB and YRGB indices, studies show that the VG range extends from -0.8V to + 0.8V. For the derivative method (I’CAT (VG)), the range of VG to discriminate is between -0.8V to -0.5V, with the maximum point of discrimination for VG = -0.65V. Finishing the studies of the indexes, the integrative method (ICAT, int (VG)) informs that the best range to work on is 0V to 0.5V, where the behavior is linear, presenting the angular coefficients (BX) of each length as discriminator RGB waveform. In the second part, the study of the algorithms generated for the characterized data of the lateral PIN gated photodiode is carried out, being: Linear Discriminant, Quadratic Discriminant, Linear SVM, Quadratic SVM, Cubic SVM, Cosine KNN and Ensemble Subspace Discriminant. All models during the training phase showed accuracy above 90% and ROC / AUC = 1. In the test phase, all models were accurate above 94%, with three of them, Linear Discriminant, Quadratic Discriminant and Quadratic SVM obtained 100% accuracy. In the third part, discrimination is proposed through 6 the modification of an electronic circuit discussed in item 3.6 of this work, through the VG polarizations which are four in total, -1.0 V, -0.6 V, +0.6 V and +1.0 V. This circuit provides the relationship between the output voltage V? by the wavelength ?, through the logarithmic ratio between the currents of the lateral photodiodes with port. The best detection response was presented for VG1 = +1.0 V and VG2 = -0.6 V, for the three optical powers in analysis (0.1; 0.5 and 1.0 W / m2). The other VG combinations show satisfactory results for any value of optical power for the ? range between 200 nm to 500nm, where there is a good variation in the output voltage (V?) of the circuit. The gate terminal proposed in this device positively assists in the discrimination of wavelengths, opening new horizons in the detection of electromagnetic radiation