<![CDATA[The need for ornamental caladium plants is very popular, but there are several obstacles to recognizing its type. Caladium species classification using AI is needed to overcome the problem of misidentification among enthusiasts. This study uses the Generative Adversarial Network (GAN) algorithm to generate new images from the Caladium dataset: Amazon Caladium, Bicolor Caladium, White Queen Caladium, and Skull Caladium. We combine GAN with YOLOv5 to detect Caladium in real time to improve accuracy. The quality of the generated images is evaluated using the Kernel Inception Distance (KID) method, with the highest scores of 0.2320 for Amazon Caladium, 0.1966 for Bicolor, 0.1713 for Skull, and 0.1857 for White Queen, indicating close similarity to the original images. We chose the best model to generate three datasets: Original Dataset, Mixed Dataset (original images plus GAN-generated images), and Dataset consisting mainly of GAN images. The Mixed Dataset achieved the best results, with a mean Average Precision (mAP) of 0.695 for an Intersection over Union (IoU) of 0.50:0.95 outperforming the GAN dataset and the original Dataset. This training used 50 epochs, a learning rate of 0.0003, and a batch size of 16, to obtain the best model and significantly improve Caladium detection. From this experiment, it was found that the GAN, combined with the original data, was able to support the accuracy of YOLOv5 for real-time caladium classification and was also able to create new images that resembled the original leaves. In the mobile application, this model allows real-time identification of Caladium types, making it easier for users to buy Caladium according to the desired type.]]>
