| As one of the world's three major economic grain crops, the yield of corn increased significantly in recent years, this has made China's grain production structure further improved. The corn production of China came to the world's second place, ranking only second to America, accounted for about one fifth of the global amount. However, because of the weakness of the processing technology after harvest, the corn quality went down during storage, and problems about the prevention and controlling in insects and microorganisms in stored corn were still severe, there were massive wastage of corn every year, especially caused by insects. The Chinese'Twelfth Five-Year Plan'about the development of grain industry technology, pointed out that one of the mian mission for grain industry during this period is the construction of six key projects, including Grain storage facility project, Grain depot and storehouse renovation project, The modern logistics for grain project, special project for grain storage by farmers, Upgrade project for grain and oil process industry, The quality of grain safety monitoring system project. Continue to carry out the scientific storage, promoting scientific storage technology, improve grain storage conditions, reduce the loss of grain after harvest. This article in view of the growing characteristics of insects in stored corn in northeast China, choosing Sitophilus zeamais, Sitophilus oryzae Linne and Tribolium confusum Jac.du Val. as research objects, using acoustic analysis, vibration analysis and pattern recognition technology, to analyze acoustic signals and vibration signals of the above three kinds of insects in stored corn during their activities as follows:1. A small silence cabinet at a size of50x45x45cm3was designed to meet the need of signal collection under laboratory conditions. Suitable sensors and DAQ cards were chosen, the device was improved to collect acoustic signals at a sampling frequency of44.lkHz, saved the data in WAV form, and vibration signals of lOOOHz saved in TXT form.2. Using a digital bandpass filter with parameters:wp=[505000]Hz, ws=[455100]Hz and hanning window to filtering acoustic signals, used coif5wavelet denoise the acoustic signals after filtering with default threshold arithmetic. Calculated signal energy in time-domain, detected endpoint of signal after pretreatment by analyzing short time frame energy in conjunction with short time frame zero-crossing rate (ZRC) to find out the important part of a signal. Analyzed the important part in frequency domain, acquired its magnitude spectrum and power spectrum to get the frequency (Hz) of each characteristic peak in proper sequence as the feature parameters of acoustic signals. Acquired Mel-scale Frequency Cepstral Coefficient (MFCC) of the important part in cepstrum domain as another feature parameters of acoustic signals.3. Using a multi-bandpass digital filter with cut-off frequency50x (2n-l)Hz (n=l,2,...,10) to filtering vibration signals multiplied with hanning window, and dbl wavelet to denoise the vibration signals after filtering with forced threshold arithmetic. Analyzed autocorrelation function of vibration signals after pretreatment in time domain, the important part in frequency domain, acquired its magnitude spectrum and power spectrum to get the frequency (Hz) of each peak in proper sequence as the feature parameters of vibration signals. Obtained duration for the important part based on Pseudo Winger-Ville Distribution in time-frequency domain analysis, this method also confirmed that there were periodical components in vibration signal which were firstly found by autocorrelation analysis. Acquired Linear Predictive Cepstral Coefficients (LPCC) of the important part in cepstrum domain as another feature parameters of vibration signals.4. Chose acoustic signal energy as an index to estimate the existence of insects at11dB, the detection limit of quantity of each insect is Sitophilus zeamais:5/kg; Sitophilus oryzae Linne5/kg and Tribolium confusum Jac.du Val.10/kg. Selected four frequency values of characteristic peaks of acoustic signal among six, which were the1st,4th,5th,6th, as the neural units of input layer for probabilistic neural network (PNN), the recognition accuracy of different insects was97.33%. Chose the1st,2nd,4th,5th frequency values of characteristic peaks of vibration signal among five as the neural units of input layer for PNN, the recognition accuracy of different insects was93.33%. Analyzed the correlation between acoustic signal and vibration signal, aquired two principal components which can represent most of the information of acoustic and vibration signals, principal components were used as inputs for PNN, the recognition accuracy of different insects was89.33%. Recognized different insects by MFCC and its first difference coefficient of acoustic signal based on HMM with a recognition accuracy about83.67%, while LPCC and its first difference coefficient of vibration signal got a recognition accuracy about77.17%. |
