Spectral Wavelength Selection Method And Its Application In Near Infrared Detection

With the progress of the times,people’s demands for product are getting higher,so not only the product testing technology is required to be high-speed and lossless,but also the highaccuracy.The traditional testing method obviously can’t meet the development needs of now.Near-infrared spectroscopy analysis technology,which takes non-destructive as its absolute advantage has emerged,and has been tested in many fields.In general,this technique is accompanied by much spectral data,which inevitably contains much noise,no information or even interfering variables,which often leads to detection model data too large and an insignificant prediction effect.Spectral wavelength screening can significantly reduce the model,speed up the prediction,improve the accuracy,and avoid over-fitting of the model with effect,which has a great impact on the development of near-infrared spectroscopy analysis technology.To solve these problems,the main research contents of this thesis are as follows:(1)Propose a variable screening method for improved Team Progress Algorithm(iTPA).iTPA divides the wavelength variables into several bands equally,and arranges the bands in descending order of the evaluation value function obtained by the Partial Least Square(PLS)modeling of their corresponding physical and chemical values.The improvement is divided into elite group,general group,garbage collection group.When the new study band chooses the learning behavior,if it occurs in the elite group,its update direction needs to be improved.In order to adjust to the opposite direction of the garbage collection group model,the evaluation value of the garbage collection group members has been extremely low along with the update process.The state can provide an accurate update direction for it,and improve the algorithm’s global optimization capability.Through continuous iterative updates,the band with the highest evaluation value is finally selected as the screening band.The results indicate that the algorithm can effectively filter out a very few of wavelengths with high accuracy,reduce the complexity,and can explain the chemical properties of each content.(2)The advantage of the iTPA algorithm is that it can filter out a very small number of wavelengths,and the prediction accuracy is not very good.Under high-precision measurement conditions,an improved Immune Genetic Algorithm(i IGA)wavelength variable screening method is proposed.Compared with the IGA and the iTPA algorithm,this algorithm is obviously Improve the prediction accuracy.Instead of the idea of fixed antibody similarity threshold in the original algorithm,this algorithm adopts adaptive antibody similarity threshold,introduces elitist retention strategy,and introduces greedy algorithm idea in the algorithm,which randomly reverses one locus for local optimization,and ensures that the affinity value of the gene after the optimization is greater than that before the optimization,and then replaces the gene after the optimization with the worst affinity in this iteration gene.The results indicate that this improved algorithm can find the optimal combination of wavelengths in the full spectrum,and it improves the model prediction accuracy compared with the original algorithm.(3)For verifying the advantage of the iTPA and the i IGA wavelength selection algorithm,while applying the algorithm to the actual detection system.The near-infrared spectroscopy data and sugar value of cherry tomatoes were successively measured for modeling and prediction.The two improved algorithms have achieved satisfactory results in experiments.Finally,a near infrared spectrum off-line prediction software system was designed to predict the content of cherry tomatoes.