An Agricultural Insect Identification System Based On Sound Signal Processing

With the development of modern agriculture, the concept of green agriculture had deeply rooted people’s mind. People are no longer just only concerned with agricultural production, but also pay attention to the safety of agricultural products. In agricultural systems, insects occupy an important part. The quality of treatment methods for insects directly affects the quality of agricultural products. In traditional agriculture, a common approach is spraying a lot of pesticides to kill all the insects that may exist, regardless of whether the insects are good for agricultural production. This extensive solution brings a variety of problems. Some of considerable funds are used to purchase the pesticides. It not only increases the cost of agricultural activity, but also take no account of whether the presence insects are pests. It just keeps spraying pesticides, resulting in a large amount of pesticide residues, meanwhile brings a serious impact on food health and the environment. Hence it is very meaningful to find an efficient way to accurately identify the insect.To solve the problem of identifying the insect sound, the paper analyzed the current situation and some problems about insect identification though referencing lager number of documents. It started from the perspective of a study of sound signal processing, introduced the processes of sound recognition and all aspects of the method. In addition, the paper also set up a comprehensive online and offline identification system. The system was made up of signal collection, processing and recognition modules. It described the configuration and data processing issues collector step by step. Considering the characteristics of insect sounds at the same time, new insect identification methods based on spectrogram were proposed. The approaches combined sound signal separation technology with improved voice activity detection (VAD) algorithm in order to extract different insect sound recognition templates which were projection in frequency direction and characteristics spectrogram respectively. Finally, experimental data show that the accurate rate of two identify methods is up to 95%. Therefore, the system has an important significance and value.