Pest attacks are one of the main problems in melon cultivation, significantly impacting productivity and crop quality. Manual pest identification has limitations in terms of objectivity, consistency, and efficiency, especially in medium to large-scale agricultural fields. This study developed a computer vision-based visual detection system for melon pests by utilizing the YOLOv9 architecture and public datasets obtained from the Roboflow platform. The dataset used consisted of 1,198 images, divided into 879 training images, 131 validation images, and 188 test images. The model training process employed data augmentation techniques, generating three outputs per training example and adding noise up to 2.52% of pixels to enhance the model's resilience to visual variations. The research methodology included system architecture design, data preprocessing, model training, and performance evaluation using precision, recall, and mean Average Precision (mAP) metrics. The test results showed that the system achieved mAP@50 of 61.6%, with 56.9% precision and 58.8% recall, indicating adequate detection capability with good inference efficiency. Thus, the developed system has the potential to be used as an early detection mechanism for melon plant pests to support decision-making in precision agriculture.