Aplikasi CNN untuk Analisis Visual Pertumbuhan Tanaman Bitter Melon dalam Sistem Akuaponik

Abstract

Technological advances in modern agriculture face major challenges, such as limited land and climate change that affect crop productivity. One approach that is gaining popularity is the aquaponic system, which is a farming method that combines fish and plants in one controlled ecosystem. In this study, a Convolutional Neural Network (CNN) method with a transfer learning approach was used, using the ResNet50 model to classify the condition of bitter melon plants growing in an aquaponic system. The developed model aims to distinguish plants into two categories, namely Good Condition and Reject. Test results show that the model has a high level of accuracy in classifying plant conditions, with a precision of 92%, recall of 100%, and F1-score reaching 96% on training data. However, the model still faces challenges in generalizing to the test data, which indicates the possibility of overfitting. To improve the performance of the model, various optimization techniques such as data augmentation and model regulation were performed to enrich the dataset variation and improve the model's ability to recognize more diverse plant growth patterns. Although there are still obstacles in handling differences in lighting and image capture angles, this method makes a significant contribution to the development of a more efficient and accurate artificial intelligence-based monitoring system in aquaponics systems. This research can be further developed by creating a more lightweight and adaptive model, and testing its performance in various real conditions in the aquaponics environment. The implementation of this deep learning-based classification system is expected to support precision agriculture innovation and encourage the sustainability of technology-based food production.