Song Meter Mini在基于未標(biāo)記數(shù)據(jù)的南非鳥類自動(dòng)生物聲學(xué)監(jiān)測(cè)中的應(yīng)用
Abstract
Analyses for biodiversity monitoring based on passive acoustic monitoring (PAM) recordings is time-consuming and chal lenged by the presence of background noise in recordings. Existing models for sound event detection (SED) worked only on certain avian species and the development of further models required labeled data. The developed framework automatically extracted labeled data from available platforms for selected avian species. The labeled data were embedded into recordings, including environmental sounds and noise, and were used to train convolutional recurrent neural network (CRNN) models. The models were evaluated on unprocessed real world data recorded in urban KwaZulu-Natal habitats. The Adapted SED-CRNN model reached a F1 score of 0.73, demonstrating its efficiency under noisy, real-world conditions. The proposed approach to automatically extract labeled data for chosen avian species enables an easy adaption of PAM to other species and habitats for future conservation projects.
摘要：
基于被動(dòng)聲學(xué)監(jiān)測(cè)（PAM）記錄的生物多樣性監(jiān)測(cè)分析是耗時(shí)的，并且受到記錄中存在背景噪聲的挑戰(zhàn)。現(xiàn)有的聲音事件檢測(cè)（SED）模型僅適用于某些鳥類，進(jìn)一步模型的開發(fā)需要標(biāo)記數(shù)據(jù)。開發(fā)的框架自動(dòng)從選定鳥類的可用平臺(tái)中提取標(biāo)記數(shù)據(jù)。標(biāo)記的數(shù)據(jù)被嵌入到錄音中，包括環(huán)境聲音和噪聲，并用于訓(xùn)練卷積遞歸神經(jīng)網(wǎng)絡(luò)（CRNN）模型。這些模型是在夸祖魯-納塔爾省城市棲息地記錄的未經(jīng)處理的現(xiàn)實(shí)世界數(shù)據(jù)上進(jìn)行評(píng)估的。Adapted SED-CRNN模型達(dá)到了0.73的F1分?jǐn)?shù)，證明了它在嘈雜的現(xiàn)實(shí)世界條件下的效率。所提出的自動(dòng)提取選定鳥類物種標(biāo)記數(shù)據(jù)的方法使PAM能夠輕松適應(yīng)其他物種和棲息地，以用于未來的保護(hù)項(xiàng)目。
關(guān)鍵詞：Song Meter，鳥鳴記錄，野生動(dòng)物聲學(xué)監(jiān)測(cè)，鳥類聲學(xué)記錄，鳥類被動(dòng)式聲學(xué)監(jiān)
