Muhammad Asyrof

Pengembangan Aplikasi Prediksi Tingkat Kematangan Buah Manggis Menggunakan Hybrid Convolutional Neural Network

Introduction

Pengembangan aplikasi prediksi tingkat kematangan buah manggis menggunakan hybrid convolutional neural network. Aplikasi web prediksi kematangan manggis dengan Hybrid CNN (InceptionV3, DenseNet201, ResNet50V2) untuk seleksi kualitas ekspor. Akurasi tinggi & AUC 0.9516.

Abstract

Penelitian ini mengembangkan aplikasi berbasis web untuk memprediksi tingkat kematangan manggis menggunakan pendekatan Hybrid Convolutional Neural Network (CNN) yang menggabungkan arsitektur InceptionV3, DenseNet201, dan ResNet50V2. Aplikasi dikembangkan dengan Vue.js untuk frontend dan Flask untuk backend mengikuti model Waterfall dari metodologi System Development Life Cycle (SDLC). Penelitian ini menggunakan dataset 1.681 gambar manggis yang diklasifikasikan ke dalam tiga kategori, yaitu Ripe, Semi-Ripe, dan Unripe. Hasil evaluasi menunjukkan performa model yang sangat baik dengan nilai AUC 0,9516 (Micro-Average) dan 0,9482 (Macro-Average). Akurasi tertinggi dicapai pada kelas Semi-Ripe (88,17%) dan Unripe (87,50%), sementara kelas Ripe mencapai 73,91%. Aplikasi ini diharapkan dapat membantu petani dan pemangku kepentingan industri dalam proses seleksi kualitas buah manggis untuk memenuhi standar ekspor.

Review

This paper presents a compelling approach to address the practical challenge of mangosteen ripeness assessment through the development of a web-based application. The core innovation lies in the utilization of a Hybrid Convolutional Neural Network (CNN) architecture, cleverly combining the strengths of InceptionV3, DenseNet201, and ResNet50V2. The application's robust development, following a Waterfall SDLC model with Vue.js for the frontend and Flask for the backend, demonstrates a systematic engineering approach. Furthermore, the use of a substantial dataset comprising 1,681 mangosteen images across three maturity categories (Ripe, Semi-Ripe, Unripe) lends credibility to the experimental setup. The reported performance metrics, notably high Micro-Average AUC of 0.9516 and Macro-Average AUC of 0.9482, suggest a highly effective model capable of accurate prediction, positioning this work as a significant contribution to precision agriculture and quality control. While the overall performance is impressive, particularly the high AUC scores, the abstract reveals an interesting nuance in the class-specific accuracies. The model achieved excellent accuracy for Semi-Ripe (88.17%) and Unripe (87.50%) categories, but a comparatively lower accuracy for the Ripe class (73.91%). This disparity warrants further investigation within the full paper, potentially exploring whether the visual characteristics of ripe mangosteens are inherently more ambiguous, or if there's an imbalance in the dataset distribution for this specific class. It would also be beneficial for the paper to elaborate on the *mechanism* of the "hybrid" CNN – how these distinct architectures are combined or leveraged – rather than just listing them, to fully understand the novelty of the proposed model. Clarification on the specific number of images per class would also provide valuable context for interpreting these results. In conclusion, this research offers a well-structured and technically sound solution for automated mangosteen ripeness prediction, leveraging advanced deep learning techniques within a practical web application framework. The high overall prediction accuracy, coupled with its accessibility, makes it a valuable tool with strong potential to assist farmers and industry stakeholders in enhancing quality control, particularly for export standards. Future work could build upon this foundation by exploring real-time deployment challenges, expanding the dataset for the Ripe category to potentially improve its accuracy, or investigating the model's generalizability to other fruit varieties. This paper represents a significant step towards intelligent agricultural systems and is a commendable effort in bridging cutting-edge AI with tangible agricultural needs.

Full Text

Comments

You need to be logged in to post a comment.