Classification of Chili Varieties Using Convolutional Neural Network Model

The diversity of chili types poses a challenge in accurately identifying chili varieties, hindering the selection of superior varieties by farmers and industry. To address this issue, our research focuses on leveraging machine learning, specifically Convolutional Neural Network (CNN) models, to classify visual images of chili varieties. We collected a dataset comprising 13 classes with a total of 1835 images, representing various species and varieties of chili. The dataset was divided into 70% for training, 20% for validation, and 10% for testing. The training process involved the use of deep learning with several CNN models to evaluate their effectiveness in accurately classifying chili varieties based on visual characteristics. Through refining the model architecture and parameters, our aim is to achieve a classification accuracy above 0.9, demonstrating the potential of CNN in overcoming current challenges in chili variety identification. In the training data experiment, models such as ResNet 50V2, NasNetMobileV2, Densenet169, MobileNetV2, Xception, and InceptionV3 demonstrated accuracy, precision, and recall of more than 92%. Subsequently, during testing data processing, the highest accuracy was achieved by the DenseNet 169 model at 95%, followed by MobileNetV2 at 95.9%, Xception at 94.3%, and InceptionV3 at 91.3%. Overall, our research has successfully classified 13 different varieties of chili peppers. However, further research is warranted to address more diverse image characteristics in other types of images.