Study On Nondestructive Detection Of Freshness Of Post-harvest Spinach Based On Machine Vision And Electronic Nose

Spinach is a common green leaf vegetable in Chinese residents’ dietary, for it is rich in nutrient elements such as minerals and vitamin. Post-harvest spinach is easy to loss water with wilting and shrinking which leads to freshness-loss, and the edible quality and product value both decreased. Therefore, there is need to study on the nondestructive method for the detection of the post-harvest spinach freshness to ensure the quality of spinach and improve its market competitiveness. This research attempt to apply machine vision and electronic nose in the detection of spinach freshness using image information and odor inforamtion.The main contents of this research are as follows:1. Study on Detection of Spinach Freshness Based on Machine Vision 1) The hardware system of machine vision which was suitable for the image acquisition of spinach was built. The type of equipments such as camera, len, lights and background color was selected, and the work of system adjustment was conducted to ensure the stability of this system.2) The discriminant analysis models to evaluate spinach freshness based on image information were researched. Leaf area of the spinach was extracted as the region of interest. Otsu method was used in the segementation of the intact spinach and the back 2/3 of the segemented region was further extracted to reduce the area to be processed. After that, morphology and regional difference set operation were used to realize the complete segmentation of leaf area. Then 18 color features variables namely (?),R_δ, G_δ, B_δ, H_δ, S_δ, V_δ, L_δ, a_δ, b_δ were extracted from the leaf image and the variables were used for the establishment of the models respectively named K-nearest neighbor and back propagation artificial neural network. The discriminant rate of K-nearest neighbor model reached 92.71% in training set and 85.42% in prediction set. While the discriminant rate of back propagation artificial neural network model equaled to 91.67% and 85.42% in training set and prediction set respectively. Both the models realized the discriminant of different freshness levels of spinach during storage. Moreover, K-nearest neighbor model performed better than back propagation artificial neural network model.3) The quantitative prediction models were established to predict chlorophyll content. Different color features variables were extracted from the leaf image. The method of backward selection was used to select the variables which were highly relevant to the chemical values of chlorophyll content. Finally, 6 color features variables namely HSHHS (?),g+b,b/g were finally chosen and were later used in the establishment of prediction models. The models named partial least squares regression analysis and back propagation artificial neural network were established. The modeling results show that, the root means square error of prediction set(RMSEP) and the correlation coefficient(Rp) were respectively 0.2315 and 0.7338 in the model of partial least squares regression analysis. While the root means square error of prediction set(RMSEP) and the correlation coefficient(Rp) of back propagation artificial neural network model were respectively 0.2147 and 0.7995. The results show that it is basically feasible to predict chlorophyll content using image information, and the back propagation artificial neural network model performs better than partial least squares regression model.2. Study on Detection of Spinach Freshness Based on Electronic Nose1) The optimization of sensor array of electronic nose was carried out. The stable values of each sensor were extracted as features variables. Loading analysis was used in array optimization. The one which has similar loading factor to the other sensors was selected as the representative of these sensors. According to the result of loading analysis, the sensors which were numberd as 9, 3, 8, 11, 1, 4, 10 finally made up the new sensor array.2) The discriminant analysis models to evaluate spinach freshness were researched. The stable values of the new sensor array were used as the features variables for the establishment of discriminant models namely support vector machine and back propagation artificial neural network. The results show that, the discriminant rates of support vector machine model equaled to 84.38% and 75.00% in training set and prediction set respectively. While the discriminant rates of back propagation artificial neural network model reached 88.54% in training set and 81.25% in prediction set, which performed better than the model of support vector machine. The results show that electronic nose can classify different spinach freshness levels during storage.3) The quantitative prediction model for chlorophyll content was established. The features variables continued to be the stable values of the new sensor array. And back propagation artificial neural network model was built as the prediction model based on electronic nose. The root means square error of calibration set(RMSEC) and the correlation coefficient(Rc) were respectively 0.3119 and 0.7013, while the root means square error of prediction set(RMSEP) and the correlation coefficient(R_p) were respectively 0.3023 and 0.6905. The results show that, the back propagation artificial neural network model can predict the chlorophyll content in some degree, but the prediction accuracy gets lower than that of machine vision.3. Study on the detection of spinach freshness based on fusion information combining machine vision and electronic nose was conducted. Image information and odor information were integrated to obtain the more comprehensive sensory information. Data fusion by combining the features variables extracted from different technology was adopted. And back propagation artificial neural network model was built for both the discriminant analysis of spinach freshness and the prediction of chlorophyll content. The model results show that, the discriminant rates of this model based on fusion information increased to 97.92% and 93.75% in training set and prediction set respectively. In addition, the predicition accuracy also improved, for the root means square error of calibration set(RMSEC) and the correlation coefficient(R_c) were respectively 0.1759 and 0.8888, while the root means square error of prediction set(RMSEP) and the correlation coefficient(R_p) were respectively 0.2121 and 0.8736.The results show that it is feasible applying machine vision and electronic nose in the detection of spinach freshness. Meanwhile, the method of data fusion is helpful in improving detection accuracy.