Identification of Arabica coffee roasting levels using Nikon D3100 commercial camera and optimized neural network

The roasting levels of coffee beans can be classified into three categories i.e. light roast, medium roast, and dark roast. The roasting level can be identified by a combination of artificial intelligence and computer vision methods. This research aimed to classify the roasting levels of Arabica coffee beans using computer vision (Nikon D3100) and an optimized artificial neural network (ANN). This study used color features i.e. red-green-blue (RGB) and hue-saturation-value (HSV), as well as textural features based on gray-level co-occurrence matrix (GLCM) including energy, contrast, correlation, and homogeneity. Features selection (ReliefF) was used to optimize ANN modeling. Sensitivity analysis was done by varying the learning function, the number of hidden layers, the activation function, and the number of nodes in the hidden layer. The best ANN architecture was obtained with a 9-30-40-1 structure (9 inputs, the first hidden layer using 30 nodes, the second hidden layer using 40 nodes, and 1 output) when using a learning rate of 0.1, a learning function of trainlm, and an activation function of logsig-logsig-purelin. The accuracy showed a mean square error (MSE) value of 0.0089 with an R-value of 0.99293 on training data, and an MSE value of 0.0115 with an R-value of 0.99143 on validation data. The test results showed an accuracy value of 97.78%. In conclusion, a combination of computer vision and ANN methods as non-destructive sensing can be used effectively to classify the roasting levels of Arabica coffee beans.