Visão computacional para classificar a maturação dos frutos de café no processo de colheita mecanizada
Data
2023-04-28
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Lavras
Resumo
O café é um dos produtos agrícolas mais comercializados e consumidos no mundo, fundamental para o desenvolvimento socioeconômico do Brasil. A colheita do café é um processo essencial na cadeia produtiva e, corresponde por aproximadamente metade dos custos totais de produção. Nesse sentido, esta pesquisa teve como objetivo classificar frutos de café quanto ao grau de maturação durante o processo de colheita mecanizada utilizando técnicas de visão computacional. Vídeos dos frutos de café colhidos foram obtidos durante o processo de colheita mecanizada na safra de 2022. A coleta de dados ocorreu sobre a espécie arábica, variedade Bourbon Amarelo, cultivada na Fazenda Cafua no município de Ijaci, localizada na região Sul de Minas Gerais. Para a coleta de imagens, foi desenvolvido um dispositivo instalado sobre a esteira transversal da colhedora, com acoplamento de uma câmera em um suporte para reduzir os efeitos da vibração da colhedora de café e com um sistema iluminação por led para a iluminação dos frutos durante a obtenção dos vídeos. Para o processamento das imagens coletadas foram realizadas duas abordagens, (i) com o desenvolvimento de um algoritmo utilizando técnicas de visão computacional e (ii) utilizando um algoritmo de detecção de objetos de última geração o YOLOv7. O algoritmo de visão computacional foi capaz de detectar e classificar frutos de café de acordo com os seguintes graus de maturação: não maduro e maduro. A precisão média para as classes de maturação do café não maduro e maduro foi de 72% e 70%. Com algoritmo não foi possível classificar os frutos da classe demasiado maduro. O algoritmo de detecção de objetos denominado YOLOv7 foi implementado para a detecção e classificação dos frutos de café em três classes: não maduro, maduro e demasiado maduro. A rede YOLOv7 apresentou capacidade superior com valores de F1-score de 90%, 95% e 75% para as classes não maduro, maduro e demasiado maduro, respectivamente. Com a classificação da maturação dos frutos de café colhidos é possível obter um índice de maturação dos frutos durante o processo de colheita mecanizada. Além disso, os resultados desse estudo podem contribuir para o desenvolvimento de sistema embarcado para ser utilizado na coleta de dados durante a colheita mecanizada do café.
Coffee is one of the most commercialized and consumed agricultural products in the world, fundamental for the socioeconomic development of Brazil. Coffee harvesting is an essential process in the production chain and accounts for approximately half of total production costs. In this sense, this research aimed to classify coffee fruits according to the degree of maturation during the mechanized harvesting process using computer vision techniques. Videos of the harvested coffee fruits were obtained during the mechanized harvesting process in the 2022 harvest. Data collection took place on the Arabica species, variety Bourbon Amarelo, cultivated at Fazenda Cafua in the municipality of Ijaci, located in the southern region of Minas Gerais. For the collection of images, a device installed on the transverse mat of the harvester was developed, with a camera attached to a support to reduce the effects of the vibration of the coffee harvester and with a LED lighting system for illuminating the fruits during harvesting. obtaining the videos. Two approaches were used to process the collected images, (i) with the development of an algorithm using computer vision techniques and (ii) using a state-of-the-art object detection algorithm, YOLOv7. The computer vision algorithm was able to detect and classify coffee fruits according to the following degrees of maturation: unripe and ripe. The average precision for the unripe and mature coffee maturation classes was 72% and 70%. With algorithm it was not possible to classify the fruits of the class too ripe. The object detection algorithm called YOLOv7 was implemented for the detection and classification of coffee fruits into three classes: unripe, ripe and overripe. The YOLOv7 network showed superior capacity with F1-score values of 90%, 95% and 75% for the unripe, mature and overripe classes, respectively. With the classification of the maturation of the harvested coffee fruits, it is possible to obtain an index of fruit maturation during the mechanized harvesting process. In addition, the results of this study can contribute to the development of an embedded system to be used in data collection during mechanized coffee harvesting.
Coffee is one of the most commercialized and consumed agricultural products in the world, fundamental for the socioeconomic development of Brazil. Coffee harvesting is an essential process in the production chain and accounts for approximately half of total production costs. In this sense, this research aimed to classify coffee fruits according to the degree of maturation during the mechanized harvesting process using computer vision techniques. Videos of the harvested coffee fruits were obtained during the mechanized harvesting process in the 2022 harvest. Data collection took place on the Arabica species, variety Bourbon Amarelo, cultivated at Fazenda Cafua in the municipality of Ijaci, located in the southern region of Minas Gerais. For the collection of images, a device installed on the transverse mat of the harvester was developed, with a camera attached to a support to reduce the effects of the vibration of the coffee harvester and with a LED lighting system for illuminating the fruits during harvesting. obtaining the videos. Two approaches were used to process the collected images, (i) with the development of an algorithm using computer vision techniques and (ii) using a state-of-the-art object detection algorithm, YOLOv7. The computer vision algorithm was able to detect and classify coffee fruits according to the following degrees of maturation: unripe and ripe. The average precision for the unripe and mature coffee maturation classes was 72% and 70%. With algorithm it was not possible to classify the fruits of the class too ripe. The object detection algorithm called YOLOv7 was implemented for the detection and classification of coffee fruits into three classes: unripe, ripe and overripe. The YOLOv7 network showed superior capacity with F1-score values of 90%, 95% and 75% for the unripe, mature and overripe classes, respectively. With the classification of the maturation of the harvested coffee fruits, it is possible to obtain an index of fruit maturation during the mechanized harvesting process. In addition, the results of this study can contribute to the development of an embedded system to be used in data collection during mechanized coffee harvesting.
Descrição
Tese de Doutorado defendida na Universidade Federal de Lavras
Palavras-chave
Maturação, Processamento de Imagens, Inteligência Artificial, Ripeness, Image processing, Artificial Intelligence
Citação
ZANELLA, Marco Antonio. Visão computacional para classificar a maturação dos frutos de café no processo de colheita mecanizada. 2023. 68 p. Tese (Doutorado em Engenharia Agrícola) –Universidade Federal de Lavras, Lavras, 2023.