<![CDATA[Green mustard (Brassica rapa var. parachinensis) is one of the important horticultural commodities in Indonesia with high economic value, but its production has declined due to leaf pest attacks such as armyworms, Plutella larvae, and aphids. Manual pest detection, which is time-consuming and prone to errors, poses a major challenge in effective early control. This research aims to develop a pest detection system on mustard greens leaves based on a website using the NASNetMobile deep learning model and sequential architecture, to provide a practical, accurate, and easily accessible solution for farmers. The research method includes the collection of 1000 images of mustard greens from the Kaggle dataset, preprocessing with augmentation and normalization, development of a CNN model with two architectures (NASNetMobile and sequential), evaluation of model performance, and implementation of a Flask-based prototype for web-based testing. The training results show that the best architecture (NASNetMobile + sequential) achieved a validation accuracy of 94% and a validation loss of 0.1160 in 14 seconds of training. Further testing using 50 new images showed an overall detection accuracy of 96%, with 100% accuracy on pest-infected leaves and 92% on pest-free leaves. The conclusion of this research indicates that the web-based detection system using the NASNetMobile and sequential models is effective in supporting pest management on green mustard plants. This system provides easy access, quick response, and high accuracy, although further development with a more diverse dataset and field testing are needed to improve reliability in real conditions across various agricultural environments.]]>
