无人系统精细识别模式的模型开发音频及测试结果评估数据

本数据集是国家重点研发计划项目“处理器宽电压弹性设计关键技术研究”中，课题四“高能效无人系统的演示系统开发”的成果之一，专为支持水下目标监测系统中的精细识别模式而构建。该模式旨在实现高准确率的多目标分类任务，通过部署8-bit量化模型于定制化的自适应AI处理器上。数据集的核心内容涵盖了两大类多分类任务所需的训练与验证数据：一是4类典型船舶辐射噪声模拟音频（货船、客船、油轮、拖船）；二是7种常见水声通信调制模式信号（2FSK、2PSK、4FSK、4PSK、8FSK、DSSS、OFDM）。所有音频数据均通过ModSignalGenerator程序模块合成，并进行了0.5dB至70dB的宽信噪比（SNR）注入，统一采用44.1kHz采样率，保存为.wav格式。此外，数据集还包含了基于AI处理器在受控水池环境下的硬件实测评估结果，验证了8-bit模型在四分类任务中90%和七分类任务中100%的分类准确率。数据集结构遵循工业界标准，包含模型训练代码（.py）、系统部署C代码（.c）及测试结果表（.xlsx）。本数据集适用于水声信号精细化识别、高精度深度学习模型开发以及边缘侧8-bit量化模型部署的研究。

This dataset is one of the achievements of Task 4 of the National Key R&D Program of China titled "Key Technologies for Wide-voltage Elastic Design of Processors", specifically constructed to support the fine-grained recognition mode in underwater target monitoring systems. This mode aims to achieve high-accuracy multi-target classification tasks by deploying 8-bit quantized models on a customized adaptive AI processor. The core content of the dataset covers training and validation data required for two categories of multi-classification tasks: first, simulated audio of 4 typical ship radiated noises (cargo ship, passenger ship, oil tanker, tugboat); second, signals of 7 common underwater acoustic communication modulation modes (2FSK, 2PSK, 4FSK, 4PSK, 8FSK, DSSS, OFDM). All audio data are synthesized via the ModSignalGenerator program module, injected with a wide range of signal-to-noise ratio (SNR) from 0.5 dB to 70 dB, uniformly sampled at 44.1 kHz, and saved in .wav format. In addition, the dataset also includes on-site hardware evaluation results obtained from the AI processor in a controlled pool environment, which verify that the 8-bit model achieves a classification accuracy of 90% in the 4-class classification task and 100% in the 7-class classification task. The dataset structure follows industry standards, and includes model training code (.py), system deployment C code (.c), and test result tables (.xlsx). This dataset is applicable to research on fine-grained recognition of underwater acoustic signals, development of high-precision deep learning models, and deployment of edge-side 8-bit quantized models.

搜集汇总

背景与挑战

背景概述

该数据集源自国家重点研发计划项目，旨在支持水下目标监测系统的精细识别模式，包含4类船舶辐射噪声和7种水声通信调制信号的合成音频数据，并提供了基于AI处理器的硬件实测评估结果，验证了8-bit量化模型在分类任务中的高准确率。数据集适用于水声信号精细化识别、深度学习模型开发及边缘侧量化模型部署的研究。