Research On Automatic Identification Of Acoustic Information Characteristics And Species Of Typical Blueberry Pests

Blueberry fruit has high nutritional value.Studies have shown that eating more blueberries has the effect of improving human immunity and preventing various diseases.Therefore,blueberry has high economic value and broad application prospects.During the cultivation of blueberries,blueberry pests seriously affect the quality and yield of blueberries,so it is necessary to prevent and control blueberry pests.The traditional method of controlling blueberry pests is to kill all possible pests by spraying a large amount of pesticides.This method is harmful to environmental protection and food health while causing waste of resources.The premise of rational use of pesticides is to clarify the types and quantities of blueberry pests.Therefore,accurately judging the types and quantities of blueberry pests will become an important issue.This paper discusses the significance of blueberry pest identification,analyzes the main blueberry vocal pests and their vocalization principles,and introduces the methods of insect sound recognition from the perspective of pattern recognition,and builds hardware for pest sound signal acquisition.The device uses MATLAB software to identify the sound signals of blueberry pests,and also analyzes the prediction method of insect population size.The wavelet noise reduction method is used to denoise the sound signal of typical blueberry pests,and the double threshold detection method is used for endpoint detection.The preprocessed sound signal improves the recognition rate of sound.After comparison of the algorithm,five typical blueberry pests were identified by GMM classifier and EM algorithm.The highest average recognition rate can reach 87%.Compared with BP neural network,GMM is more suitable for the recognition of typical blueberry pest sound.The feature extraction algorithm is improved to adjust the framing link of the blueberry typical pest sound signal to the triangular filter bank for filtering,that is,the MFCC_E method,the highest average recognition rate of using MFCC_E can reach 90%.And the recognition time is shortened,and the pre-emphasis stage can be directly implemented by hardware by improveing pre-emphasis parameters.Use MATLAB software to build a GUI interface for the recording,training,recognition and display the recognition results of blueberry pest sound.The method of predicting the size of insect population was analyzed.The size of bread bug population was predicted by marker recapture method and de-sampling method.The marker recapture method showed better performance.