PEMANFAATAN CITRA DIGITAL DAN ALGORITMA CNN UNTUK MEMPREDIKSI WAKTU PANEN KACANG TANAH YANG PALING TEPAT

Abstract

Determining the right harvest time is crucial for maintaining the quality of peanut yields. However, farmers often face difficulties in determining harvest status quickly and accurately. This study aims to develop an Android-based application for detecting optimal peanut harvest time using digital images to support farmer decision-making. The implemented method is a Convolutional Neural Network (CNN) with the ResNet50 architecture. The research dataset consists of 510 digital images classified into three categories (Not Ready for Harvest, Ready for Harvest, and Late Harvest), divided into 70% training data, 15% validation data, and 15% testing data. The test results show that the ResNet50 model is able to achieve an overall accuracy of 90%. Model performance is supported by an average precision value of 0.91, recall of 0.89, and F1-score of 0.89. In addition, the Area Under Curve (AUC) value for each class is very high, ranging from 0.98 to 1.00. Although the 5-Fold Cross Validation results produced a lower average accuracy (43%), the main test results proved that the CNN-ResNet50 architecture is effective and has the potential to be an accurate tool for farmers in determining the optimal peanut harvest time.