Kuantifikasi Sinyal Akustik Target Buatan Menggunakan Prototipe Sonar dan Algoritma Convolutional Neural Network

Abstract

Penelitian ini bertujuan mengukur nilai Target Strength (TS) material kayu, PVC, dan besi menggunakan prototipe sonar berbasis PZT frekuensi 20 kHz serta mengevaluasi klasifikasi algoritma Convolutional Neural Network (CNN). Eksperimen dilakukan pada kedalaman 3 meter di kolam terkontrol. Hasil pengukuran TS menunjukkan nilai -40,56 ± 2,469 dB (kayu), -35,8 ± 1,718 dB (PVC), dan -41,88 ± 1,874 dB (besi). Evaluasi model CNN pada 399 sampel data membuktikan bahwa penggunaan data Continuous Wavelet Transform (CWT) memberikan hasil jauh lebih optimal dibandingkan data mentah (raw data). Model dengan data mentah menghasilkan akurasi 0,83 dan AUC 0,85, sedangkan model berbasis CWT mencapai akurasi sempurna dengan AUC 1,00 dan F1-Score 1,00. Disimpulkan bahwa integrasi teknologi akustik dengan CNN melalui pemrosesan sinyal CWT menunjukkan potensi besar untuk identifikasi dan klasifikasi target bawah air, meskipun hasil ini masih terbatas pada kondisi kolam ideal dengan potensi overfitting.

This study aims to measure the Target Strength (TS) value of wood, PVC, and metal materials using a 20 kHz PZT-based sonar prototype and evaluate the classification of the Convolutional Neural Network (CNN) algorithm. The experiment was conducted at a depth of 3 meters in a controlled pool. The TS measurement results showed values of -40,56 ± 2,469 dB (wood), -35,8 ± 1,718 dB (PVC), and -41,88 ± 1,874 dB (metal). Evaluation of the CNN model on 399 data samples proved that the use of Continuous Wavelet Transform (CWT) data provided much more optimal results than raw data. The model with raw data produced an accuracy of 0,83 and an AUC of 0.85, while the CWT-based model achieved perfect accuracy with an AUC of 1.00 and an F1-Score of 1.00. It was concluded that the integration of acoustic technology with CNN through CWT signal processing shows great potential for underwater target identification and classification, although these results are still limited to ideal pool conditions with the potential for overfitting.