Study On Intelligent Detection Method Of Tomato Ripeness During Storage

Tomatoes are not only delicious with sour and sweet taste,but also can provide the vitamins and mineral salts needed by human body.They are loved by the people,but can not be stored for a long time after ripening.In order to prolong the storage time,tomatoes are generally picked at the earlier stage of their growth.During the storage,tomatoes will continue to turn ripe.Tomatoes with higher maturity not only has a poor storage performance,but also induce the surrounding tomatoes entering early into the respiratory climacteric stage,and prone to corruption.Therefore,it is of great significance to develop an objective,intelligent and non-destructive method to detect the ripeness of tomatoes.This research attempts to apply machine vision,electronic nose and hyperspectral sensing in evaluating the ripening of tomatoes.The main contents of this research are as follows.1.Study on detection method of tomato ripeness using machine vision(1)The hardware system of machine vision was built for tomato ripeness detection.The image processing algorithm for tomato detection was developed.Taking tomato as the research object,the components of hardware system such as illumination source,camera and lens were selected and/or designed.The automatic threshold segmentation method was used to remove the background of tomato images,and 6 color features[G/B/H/S/L/a*]were extracted as effective characteristic variables.(2)Study on the detection method of tomato ripeness based on machine vision.The Fisher discriminant analysis model and the support vector machine model for detecting tomato ripeness were established based on the color features of tomatoes,respectively.The results showed that the two models got the same recognition accuracy,90.28%for the training set and 88.89%for the prediction set respectively.(3)Study on the prediction method for related indicators of tomato ripeness based on machine vision.The partial least squares model and the support vector machine model for predicting hardness and lycopene were established based on the color features of tomatoes,respectively.The results showed that the prediction performance of support vector machine model was better than partial least squares model.In the color feature-based hardness support vector machine model,the correlation coefficient r_c in the training set was 0.9341,and the root mean square error of cross-validation(RMSECV)in the training set was 0.0069,and the correlation coefficient r_p in the prediction set was 0.8941,and the root mean square error in the prediction set(RMSEP)was 0.0101.For lycopene support vector machine mode,r_c=0.9214,RMSECV=0.0079,r_p=0.8905 and RMSEP=0.0105.2.Study on detection method of tomato ripeness using electronic nose(1)Study on the detection method of tomato ripeness based on electronic nose.The stable response values of the tomato samples from each sensor were selected as odor features.The K-nearest neighbor model and the support vector machine model for detecting tomato ripeness were established respectively.The results showed that the recognition accuracy in the training set and prediction set of support vector machine model were 84.72%and 83.33%respectively.The recognition effect was obviously better than the K-nearest neighbor model.(2)Study on the prediction method of related indicators of tomato ripeness based on electronic nose.The partial least squares model and the support vector machine model of predicting hardness and lycopene were established respectively,based on the odor features of tomatoes.In the odor feature-based hardness support vector machine model,r_c=0.8808,RMSECV=0.0109,r_p=0.7839,RMSEP=0.0228.The prediction performance was better than partial least squares model.In the lycopene support vector machine mode,r_c=0.9304,RMSECV=0.0073,r_p=0.7954,RMSEP=0.0309.3.Study on detection method of tomato ripeness using fusion technique combing machine vision and electronic nose(1)The information fusion method of machine vision and electronic nose was established,and the information was fused at the level of feature layer.Six color characteristic variables selected from machine vision and ten odor characteristic variables selected from the electronic nose,were integrated to obtain the fusion features which contained visual information and olfactory information.(2)Study on the detection method of tomato ripeness using fusion technique combing machine vision and electronic nose.The support vector machine model of detecting tomato ripeness was established based on the fusion features of tomatoes.The results showed that the recognition accuracy were 84.72%for the training set and83.33%for the prediction set.(3)Study on the prediction method of related indicators of tomato ripeness using fusion technique combing machine vision and electronic nose.The support vector machine model of detecting hardness and lycopene were established based on the fusion features of tomatoes,respectively.The results showed that,in the hardness prediction model,r_c=0.9220,RMSECV=0.0081,r_p=0.9143,RMSEP=0.0081.In the lycopene prediction model,r_c=0.9318,RMSECV=0.0054,r_p=0.9020,RMSEP=0.0272.4.Study on detection method of tomato ripeness using hyperspectral sensing(1)Study on the detection method of tomato ripeness based on hyperspectral sensing.The hyperspectral reflectance data of tomato samples were preprocessed with Standard Normal Variable transform.The support vector machine models for detecting tomato ripeness were established,based on the full spectral variables and109 spectral variables selected by Synergy Interval-Partial Least Squares respectively.The results showed that,the latter achieved the same recognition accuracy 94.44%for the prediction set,with less spectral variables.(2)Study on the prediction method of related indicators of tomato ripeness based on hyperspectral sensing.Combined with support vector machine model,the full spectral variables and 149 spectral variables selected by genetic algorithm were used to predict hardness,respectively.The full spectral variables and 18 spectral variables selected by genetic algorithm were also used to predict lycopene respectively.The results showed that,the model established by the filtered spectral variables achieved better predictions than the full spectral model,with fewer spectral variables.In the hardness prediction model,r_c=0.8923,RMSECV=0.0055,r_p=0.8895,RMSEP=0.0054.In the lycopene prediction model,r_c=0.9133,RMSECV=0.0088,r_p=0.9120,RMSEP=0.0088.