Proceedings of The 2nd World Conference on Engineering and Technology
Coffee Moisture Classifier Using Convolutional Neural Networks
Yurley Tovar Martínez, Andrés Calvo Salcedo and Arley Bejarano Martínez
One of the most important agricultural sectors in Colombia is the coffee industry. The coffee provides important income to the country and is highly recognized for its flavor and quality. To ensure quality production, the drying process is one of the most important factors. This process requires the humidity of the production be in the range of 10% to 12%. Due to the low impact of technification process in the country, many coffee growers measure empirically this characteristic because specialized equipment tends to be highly expensive or unknown. In addition, generating inaccurate estimation of coffee moisture could produce economic losses, due to the cost overrun for hiring an experts, or the loss of weight because of the coffee dehydration, which is fundamental in the sale of coffee. In this work, we developed an algorithm that detects coffee moisture using digital images. Our method uses deep learning from a convolutional network with LeNet CNN topology to classify the image and obtain the measurement. This proposal was validated with images captured by the coffee growers themselves and finally, the results shows the implemented method is robust to outliers and changes in the capture of the photographs.
keywords: Convolutional neural network, machine learning, Support Vector Machine (SVM).