Coffee Grind Size Detection By Using Convolutional Neural Network (CNN) Architecture

Abstract

Identification of coffee grinder results is one of the needs to support the government and coffee shop UMKM to innovate coffee drink products. Special expertise and sufficient time are needed to process the identification of coffee grinder results in the laboratory. Several previous research methodologies, the process of identifying the results of coffee grinders is still manual with human visuals. While the use of a computer system is obtained from cross-sectional images of coffee grounds using microscopic and macroscopic processes. Currently, computer vision and machine learning technologies have been developed to identify various types of objects, one of which is coffee objects. This study contributes in classifying several classifications of coffee grinder results using Convolutional Neural Networks (CNN). The novelty of this research lies in improvising the optimal CNN parameters in detecting objects from a coffee grinder. The proposed AlexNet architecture has seven layers, namely three convolution layers, two max-pooling layers and two Hidden Layers for the five characters of the coffee grinder image dataset.dataset private resulting from a coffee grinder of 1039 items and an augmentation process to make it more optimal and prevent overfitting , the test first changes the input image to 50 x 50, 100 x 100, and 150 x 150 pixels and each repeats at 250 epochs. The experimental results show that AlexNet with parameters batchsize 8, Learning Rate 0.001, Optimizer SGD, training and splitting data ratio 0.6:0.4 and balanced data typehasan accuracy validation value reaching 95%.