Detection And Identification Of Wheat Pest Particles Based On Collision Acoustic Signal Processing

The insects inside the wheat kernels will damage lots of stored grain, and several potential hazards, such as the nutritional losses due to insect infestation, as well as the contamination caused by excrement and fragments, cannot be neglected. Long-term feeding on wheat damaged by insects will result in malnutrition, even induce diseases. Therefore, the work of detection and recognition of insect-damaged wheat kernels (IDK) is of great urgency.The main objects of the study in this paper are undamaged wheat kernels and insect-damaged ones. In order to implement the effective detection and recognition of IDK, three approaches based on impact acoustical signal processing, which include the method based on ensemble empirical mode decomposition (EEMD), the method based on bispectrum and integrated bispectrum analysis, as well as the method combing an improved extreme learning machine with dragonfly algorithm, were proposed in this paper for detection of undamaged wheat kernels and IDK. With the experiments, good detection accuracies were acquired from these three methods, which indicated the effectiveness of the proposed method for detection and recognition of IDK, also it can provide the basis for detection of wheat kernels and other agricultural products.The contents arranged in this paper are as follows:(1) Introduce the background and significance of the study, and summarize the detection technologies for IDK in prior work, and the research developments of the methods based on impact acoustical signals in aspects of recognition and detection of agricultural products, such as the wheat kernels, pistachios, hazelnuts, walnuts, beans, potatoes, rice, corns and so forth.(2) Describe experiment apparatus for collection of impact acoustical signals of wheat kernels.(3) For detection of IDK, the traditional methods for feature extraction in time-domain and frequency-domain in the prior work were improved in this paper, and the method of the impact acoustical signal processing based on EEMD was proposed for detection of IDK. Intrinsic mode functions (IMFs) were obtained through EEMD, and the novel and effective discriminant features were obtained, which included the values of kurtosis in the first 6 IMFs, the form factors in the first 6 IMFs, the Renyi entropies in the first 8 IMFs, as well as the mean value of the degree of stationarity, then Libsvm was used for classification, and the kernel parameter of radial basis function (RBF) as well as the penalty factor were optimized with the grid-research algorithm by finding the one giving the highest ten-folds cross-validation (CV) accuracy of the training set. With the experiments,98.7%of undamaged wheat kernels and 93.3%of insect-damaged ones were correctly detected. Compared with the prior work, the classification accuracies for two types of wheat kernels were obviously improved, which indicated that it is effective to detect IDK by using the proposed method. In addition, when 300 mildew-damaged and 300 sprout-damaged wheat kernels were added, the proposed method can still detect IDK to some extent, which further indicated the effectiveness of the proposed method for detection and recognition of IDK.(4) For detection of IDK, the traditional methods for feature extraction in time-domain and frequency-domain in the prior work were improved in this paper, and the method of the impact acoustical signal processing based on bispectrum and integrated bispectrum analysis was proposed for detection of IDK.364 features, including 54 features from the diagonal slice of bispectrum,54 features from the horizontal slice of bispectrum,128 features from the integrated bispectrum, and 128 features from the surrounding-line integral bispectrum (SLIB), were extracted as the initial features. Then principal component analysis (PCA) was adopted for further extraction of required discriminant features, Libsvm was used for classification, the kernel parameter of radial basis function and the penalty factor were optimized with the grid-research algorithm by finding the one giving the highest ten-folds cross-validation accuracy of the training set. With the experiments,92% of undamaged wheat kernels and 91% of IDK were correctly detected, which indicated the effectiveness of the proposed method for detection and recognition of IDK. In addition, compared with the prior work, the proposed method has the robustness for Gaussian noises.(5) The method combining an improved extreme learning machine with dragonfly algorithm (DA) was proposed for detection of IDK. Each time-frequency bin of the impact acoustical signals was acquired through the short-time Fourier transform (STFT), and subsequently Gaussian modeling and parameters estimation were used for each time-frequency bin. The 26 estimated parameter features, including 13 mean features and 13 variance features, were used as the discriminant features, then an improved extreme learning machine called COAS-ELM were used for recognition and detection of IDK, and dragonfly algorithm were applied for corresponding parameter optimization of the COAS-ELM. With the experiments,99.0% of undamaged wheat kernels and 97.0% of insect-damaged ones were correctly detected. Meanwhile, COAS-ELM has the capacity of fast-learning. Compared with the prior work, the proposed method not only can acquire the higher recognition rates, but also has the high detection speed. Therefore, it is effective to detect and recognize IDK by using the proposed method.