基于音频分类的森林盗伐事件场景识别

Scene Recognition of Forest Piracy Based on Audio Classification

森林是地球上最宝贵的资源,但森林和古树被盗伐的事件却经常发生。将音频分类技术应用于森林保护,通过对盗伐过程中产生的音频进行识别,达到森林保护预警的目的。针对盗伐事件的音频场景特性,提取电锯声、引擎声、机器轰鸣声、手锯声、风噪等五种声音的音频特征,然后使用支持向量机作为分类器对音频事件进行识别。最后根据识别结果确定是否有盗伐行为发生。通过不同训练样本数量对比实验、随机样本实验和单项音频识别准确率实验最终确定总体识别率为99.1%,各项音频单独识别率在90%以上,说明此方法具有较好的识别效果。

Forests are the most precious resources on the earth,but the illegal logging of forests and old trees often occurs.To this end,we applied the audio classification technology to forest protection.The purpose of forest protection early warning is achieved by identifying the audio generated during the process of illegally chopping trees.Firstly,the audio features of five kinds of sounds,such as chainsaw sound,engine sound,machine roar,hand saw sound,and wind noise,were extracted according to the audio scene characteristics of the illegal logging event.The support vector machine was then used as a classifier to identify the audio event.Finally,it was determined whether there was any illegal logging behavior based on the recognition results.The overall audio recognition rate is 99.1%,through the different training sample quantity comparison experiments,random sample experiments and single audio recognition accuracy experiments and the individual audio recognition rate is above 90%,which both indicate that the method has a better effect on the audio recognition.