Research On The Key Algorithms Of Potato Infrared Spectral Database System

Infrared spectrum is widely applied in multiple fields,including analysis of chemicals,prediction of material varieties and quality identification because of its high stability,no requirement of chemical treatment for samples,abundant map information and other advantages.Infrared spectral database system is helpful in the construction of stable and rapid sample variety prediction model,quality analysis model and characteristic analysis model etc.,and facilitates researchers to grasp sample information more comprehensively.In infrared spectral database system,accurate and efficient spectral classification matching algorithm is the key guarantee for the effective operation of the whole system.Therefore,the study on spectral classification matching algorithm can promote the development of infrared spectral database system.Current spectral matching algorithm researches consider similarity degree between spectra mainly from the aspect of Euclidean distance measure or curve similarity,but are unable to combine the two factors.In addition,with the increasing total number of sample categories,the accuracy rate of current algorithm is unable to satisfy the request of the spectral database system.Therefore,this thesis investigates the key algorithms in infrared spectral database system,and potatoes are taken as research object.Initially,considering that traditional peak recognition algorithm for spectral feature requires multiple scans of the spectrum,and that recognition abilities of small peak and shoulder peak are weak,a characteristic peak recognition algorithm based on dynamic peak shape factor which only requires one scan is proposed.It is demonstrated that this algorithm is capable of accurately recognizing all effective characteristic peaks in the curve of spectrum and has a certain ability to recognize shoulder peak and small peak.Secondly,according to the concept of the Hamming distance and the spectral curves of the difference,a dynamic spectral distance algorithm is proposed,which comprehensively considers the waveform factor and absolute difference factor of spectral curve and realizes accurate identification of different potato varieties,revealing that the average accuracy rate of this algorithm has reached 92.85%which is higher than that in traditional Euclidean distance and spectral angle algorithm,etc.Finally,as for decreased accuracy rate in spectral classification algorithm with increasing total number of class centers,a spectral classification algorithm based on virtual competitive self-organizing and self-growth feature map neural network(VC-TGSOM)is studied.Our study indicates that the accurate rate of the VC-TGSOM network doesn't decrease with increasing total number of class centers.