Dandan Ye

Pixel based bruise region extraction of apple using Vis-NIR hyperspectral imaging

Abstract Bruises on apples will directly influence its preservation and marketing for they can cause the internal decomposition and flaws of the appearance of apples. Therefore, an effective pixel based bruise region extraction method was proposed in this study to obtain the complete bruise region. Hyperspectral images of 60 apples were obtained via the hyperspectral imaging (HSI) system at 0, 12 and 18 h after the damage experiment. Principal Component Analysis (PCA) was used to compression data size and eliminating redundant data of hyperspectral image cubes. After the selection of the region of interest (ROI) by certain rules, different pixel based apple bruise extraction models were built and compared. The result shows that Random Forest (RF) model have a high and stable classification accuracy, which turns out that RF algorithm is more suitable for classifying bruises on apples than others. The average accuracy of bruise extraction models reached 99.9%. Compared with the most used image processing method in recent literature for extracting bruises of apples, the bruising region predicted by RF model was more consistent with the true bruise region. Additionally, two characteristic wavebands around 675 nm and 960 nm related to the bruise region were singled out for reducing the dimensionality of data by analyzing the feature importance scores of the built RF model. The overall results indicated that the proposed method has a great potential to detect complete bruise region on apples based on hyperspectral imaging for improving the efficiency of apple grading and sorting.

Non-destructive prediction of protein content in wheat using NIRS

A steady and accurate model used for quality detection depends on precise data and appropriate analytical methods. In this study, the authors applied partial least square regression (PLSR) to construct a model based on the spectral data measured to predict the protein content in wheat, and proposed a new method, global search method, to select PLSR components. In order to select representative and universal samples for modeling, Monte Carlo cross validation (MCCV) was proposed as a tool to detect outliers, and identified 4 outlier samples. Additionally, improved simulated annealing (ISA) combined with PLSR was employed to select most effective variables from spectral data, the datas dimensionality reduced from 100 to 57, and the standard error of prediction (SEP) decreased from 0.0716 to 0.0565 for prediction set, as well as the correlation coefficients (R2) between the predicted and actual protein content of wheat increased from 0.9989 to 0.9994. In order to reduce the dimensionality of the data further, successive projections algorithm (SPA) was then used, the combination of these two methods was called ISA-SPA. The results indicated that calibration model built using ISA-SPA on 14 effective variables achieved the optimal performance for prediction of protein content in wheat comparing with other developed PLSR models (ISA or SPA) by comprehensively considering the accuracy, robustness, and complexity of models. The coefficient of determination increased to 0.9986 and the SEP decreased to 0.0528, respectively.

The feasibility of early detection and grading of apple bruises using hyperspectral imaging: Early detection and grading of apple bruises

The ability to determine if an apple is bruised and to provide quantitative and objective descriptions of the degree of bruising is not the only important basis for assessing apple quality but also has significance for improving the postharvest handling of apples. In this study, segmented principal component analysis for hyperspectral images in the spectral range of 401 to 1037 nm was carried out, and seven characteristic wavelengths were selected based on the weight coefficients of the principal component images. By using the principal component analysis operations with the selected wavelengths and image processing methods, an accurate recognition algorithm for apple bruises was proposed. For 40 intact samples and 160 bruised samples, the average correct recognition rate was 99.1%. Moreover, this paper obtained the average spectra of 157 segmented bruised regions by applying a binary mask. A characteristic wavelength selection method that combines competitive adaptive reweighted sampling with correlation coefficient methods and supports vector machine modeling methods based on grid parameter optimization was put forward for the classification and identification of the bruising degrees of apples. The results showed that the classification accuracy was as high as 97.5% for the test set. Overall, this study demonstrated that hyperspectral imaging technology can be used to accurately and effectively identify early bruises and determine the bruising degree of apples, which provides a new method for on‐line, nondestructive detection, and grading of early bruises in apples.

Determination of bruised potatoes by GLCM based on hyperspectral imaging technique

A texture recognition method based on hyperspectral imaging technology was proposed for the difficulty in bruise detection of potatoes. Firstly, the hyperspectral images of healthy and bruised potatoes were collected, and then they were used to generate principal component images by using PCA method. To enhance the features of bruised region in these principal component images, the histogram equalization, mean smoothing, and gradient method were used, respectively, which proved that the histogram equalization method was the most suitable. Extra, through the selection of parameters of gray level co-occurrence matrix (gray level L and distance D), the best prediction results were obtained when L=8 and D=8. Finally, SA method was applied to reduce the dimension of the texture data, and the recognition rate reached 93.75%.

Extraction of characteristic spectral bands of wet gluten in wheat based on NIR

Wheat is one of the most important food crops. The content of wet gluten in wheat is a key factor to determine the degree of flour gluten. Near infrared spectroscopy was used to predict wet gluten content of wheat by establishing prediction model in this paper. The authors did some processing on the identification of abnormal samples, pretreat spectral data and partitioned calibration set to improve the predictive ability of the model, especially made a deep research on the characteristic spectral bands. The authors collected 100 spectral points and divided them into 25 groups. Through eliminating a set of spectral points to create the partial least-squares regression model and retaining the spectral combinations which had better predictive ability to continue filtering, the authors got the characteristic spectral bands. The result showed that r, R2, RPD and SEP of the model created by the whole spectral data reached 0.923, 0.848, 2.564 and 1.421 respectively, while the results of model created by the characteristic spectrum were 0.950, 0.901, 3.177 and 1.149 respectively. The predicting ability of the latter obviously improved.

Classification of wheat grains in different quality categories by near infrared spectroscopy and support vector machine

For the purpose of rapid, simple and accurate identification of quality of wheat grains, this study proposed a recognition method which is an integration of near infrared spectroscopy and support vector machine (SVM). The spectral data of wheat samples were analyzed in order to eliminate abnormal data, and then Mahalanobis distance method was used to identify abnormal samples. After deleting those abnormal samples, principal component analysis was done to prove the feasibility of classifying wheat by near infrared technologies. The remaining 111 wheat samples were divided into calibration set and prediction set by sample set partitioning based on joint X-Y distance algorithm, then, the first derivative, second derivative, standard normal variate (SNV) transformation and their combinations were used to preprocess spectra for obtaining the optimal pretreatment method before modeling. Finally, SVM and back propagation neural network classification model were established with the spectral data preprocessed by second derivative plus SNV and first derivative plus SNV, respectively. Prediction results of SVM model showed that the recognition accuracy rate of strong gluten wheat and weak gluten wheat both achieved 100% and the recognition accuracy rate of medium gluten wheat also reached 81.82%, which proved that SVM classification model with the spectra data preprocessed by the second derivative plus SNV achieved the best results and realized rapid and accurate identification and classification of wheat quality.