| Grain is necessity in human’s production and live. After grain is harvested, it will suffer damage by mold, pest and rodents etc in the process of storing. Therefore, quality of food will decline and directly endanger human body health.In the factors causing granary grain loss, pest is the foremost factor. In the world, eight percents grain in the granary loses each year, pests cause five percents loss. Therefore, early detecting and finding the amount and type of stored grain pests in granary are convenient to take targeted measures to prevent and control of pests. This is of great significance for safe grain storage.Through detecting acoustic signals of pests in activity, acoustic detection technology can detect stored grain pests in grain heaps rapidly, simply, nondestructivly and flexibly. It is a research hotspot in stored grain pest detection technology in recent years.In this paper, the research content mainly includes the following aspects:(1) It introduces the background and meaning of detecting stored grain pests" acoustic signals and summarizes the research status of stored grain pests detection technology. The research development of sound detection technology in detection of stored grain pests is summarized importantly.(2) Based on the form and behavior characteristics of Sitophilus zeamais and Tribolium castaneum. acoustic signal acquisition system of stored grain pests is designed. The device of acquisition system, recording software, and process of acoustic signal acquisition are set. Creeping and vibratory signals of Sitophilus zeamais and Tribolium castaneum were collected and all characteristics of the four active acoustic signals were analysed and contrasted.(3) Denoising experiments on stored grain pests’activities acoustic signal are proceeded. EMD. wavelet threshold de-noising, neural network adaptive filter and FastICA algorithm are interpretated respectively.These four algorithms can denoise for signals with noise. The denoising effects of each method were compared. The results show that the denoising effects of EMD and FastICA are better than other two methods in stored grain pests active acoustic signals denoising. (4) Identify pests’ acoustic signal based on GMM and clustering approach. Basic principle and method of parameters estimation of GMM, MFCC, clustering algorithm theory are introduced. The method of combinating GMM and clustering to recognize acoustic signals is expounded. The MFCC characteristics of acoustic signals are extracted first, then the GMM of MFCC characteristics data is built. Clustering algorithm can identify four kinds of active acoustic signals and all of the recognition rates achieve above84%.(5) Identify and separate of acoustic signals based on the FastICA algorithm. Firstly, FastICA algorithm is introduced. Then through separating the independent components from the blind source, mixed signals of stored grain pests are separated to independent components. The results show that independent components in mix signals are separated and the characteristics of acoustic signals are kept.This method provides solution and idea for separating mixed acoustic signals in the actual granary.(6) Identify stored grain pests’active acoustic signals based on ISOMAP and SVM method. Manifold learning method and isometric mapping algorithm are introduced. By this method, the grain insect activity sound signals are reduced from high dimensional to low dimensional, then the characteristics of ISOMAP are extracted. In addition, support vector machine theory is introduced. With SVM, optimal hyperplane can be gained and classify ISOMAP manifold features of acoustic signals. Experiment verified that different kernel function parameters can influence the optimal classification face. When the values of parameters a and b respectively are2and0.25, the influence of penalty parameter on recognition rate of SVM is discussed. When the value of parameter C is21. the recognition rate achieve a sense of stability, and the effect of classifying of signals is the stablest. Under the above three parameters, four kinds of active acoustic signals are classified. The recognition effect is well. |
