Green Bean Classification: Fully Convolutional Neural Network with Adam Optimization

The main issue faced by coffee growers is the lack of accuracy in manual sorting.Green Bean photos are classified into three categories (dark, green, light) using the convolutional neural network approach.This study employs 360 datasets, each consisting of 120 data points per class.The CNN model uses a 512×512×3 image as its input.The number three represents the use of three specific channels: red, green, and blue.RGB is an acronym that represents the colors Red, Green, and Blue.During the feature learning phase, the input image is subjected to conversion and pooling processes.The convolutions exhibit variations in the utilization of filters and kernel sizes.The model was developed utilizing the relay activation function and pooling techniques.The dropout technique entails transforming the feature maps derived from the pooling layer into a vector shape by flattening them.This study utilized three data scenarios: 70:30, 80:20, and 90:10.The number of epochs used for each scenario was 30.This analysis compared the performance of three optimization algorithms: Adam, RMSprop, and Nadam.The optimization model using Adam with a parameter of Epoch 30 and a data scenario of 80:20 achieved the highest accuracy result of 81.67%.