Convolutional Neural Network Technique for Distinguishing Nine Varieties of Vegetable Crops

초록

In the field of agriculture, conducting research on neural network models for image classification is necessary to accurately categorize crops based on their types and health conditions and distinguish them from other species, to minimize crop losses. This study aimed to compare multiple neural network models to select the optimal model that can classify the images of nine vegetable seedlings, such as carrot, Kimchi cabbage, kohlrabi, lettuce, mallow, mustard, pak-choi, spinach, and sweet pepper. The best model was selected based on its accuracy (precision, recall, and F1 score) from eight trained models, namely DenseNet201, InceptionResNetV2, InceptionV3, MobileNetV2, ResNet152V2, VGG16, VGG19, and Xception. To train the models, a 9-class dataset, 20 epochs, 32 batch sizes, Adam optimizer, and a learning rate of 0.001 were used. The DenseNet201 model exhibited the highest accuracy and was, therefore, selected as the optimal model. With a batch size of 128, Adam optimizer, and a learning rate of 0.001, this model exhibited high precision, recall, and F1 score, and its superiority was confirmed using a confusion matrix. As a result, the DenseNet201 model is expected to improve the recognition performance of the model by using images of various plant species, exploring more networks, and optimizing the hyperparameters to achieve higher recognition accuracy.

키워드