Search

In order to conquer the difficulty of simultaneously recognizing the multiple defects of potatoes samples placed randomly, this paper proposed a non-destructive detection method which combined manifold learning dimension reduction algorithm based on hyperspectral information fusion and extreme learning machine (ELM) to simultaneously distinguish the multiple defects of potatoes. In this paper, 367 potatoes were picked which were made up of 111 sprouting potatoes, 90 green rind potatoes, 46 blackheart potatoes and 120 normal potatoes. The hyperspectral image acquisition system contained imaging spectroradiometer (SPECIM, V10E, Finland), data acquisition box, lighting system, electric moving stage and objective table. The reflection hyperspectral information of all those potatoes was acquired by using the hyperspectral image acquisition system, whose spectral wavelength ranged from 390 to 1040 nm. The hyperspectral information included the spectral information from 520 wave bands and the imaging information from 520 gray images. After the correction to hyperspectral data with the standard black and white board, hyperspectral data in the range of 450-990 nm, which had high signal-to-noise ratio (SNR), were selected as original spectrum for subsequent processing. To deal with the spectral information, the average spectrum was abstracted from the region of interests (ROI) on every potato sample by using the environment for visualizing images (ENVI). Comparing several data preprocessing methods, detrend was determined as the optimal spectral preprocessing method. Diffusion maps (DM), locally linear embedding (LLE) and hessian locally linear embedding (HLLE) were respectively utilized for the purpose of cutting down the dimension of spectrum data after the spectral preprocessing named detrend. To deal with the hyperspectral imaging information, every pseudo-color image of potatoes was morphologically processed before extracting 84 image texture characteristics based on gray level co-occurrence matrix (GLCM). By using successive projections algorithm (SPA), 10 texture features were properly selected, which consisted of homogeneity of 45° GLCM in R gray image, energy of 90° GLCM in R gray image, correlation of 90° GLCM in R gray image, contrast of 135° GLCM in R gray image, correlation of 0° GLCM in G gray image, contrast of 0° GLCM in G gray image, correlation of 90° GLCM in G gray image, homogeneity of 45° GLCM in B gray image, correlation of 90° GLCM in B gray image and contrast of 135° GLCM in B gray image. After the fusion of the spectral characteristics and image features, the 367 potato samples were divided into calibration set and test set. The calibration set had 250 potato samples including 74 sprouting potatoes, 59 green rind potatoes, 28 blackheart potatoes and 89 normal potatoes. The whole number of samples in test set was 117 which contained 37 sprouting potatoes, 31 green rind potatoes, 18 blackheart potatoes and 31 normal potatoes. Six models were respectively established based on support vector machine (SVM) and ELM, which were DM-SVM, LLE-SVM, HLLE-SVM, DM-ELM, LLE-ELM and HLLE-ELM. Comparing and analyzing the results of the 6 models and the spending time, we discovered that the DM was the best manifold learning dimension reduction algorithm to deduce the spectral information. We alsofound that the DM-ELM model was the best model because of the better recognition rate for recognizing the multiple defects of potatoes and spending less time. For the DM-ELM model, the single recognition rate of sprouting potatoes, green rind potatoes, blackheart potatoes and normal potatoes respectively reached 97.30%, 93.55%, 94.44% and 100%, and the mixed recognition rate reached 96.58%. It only cost 0.11 s to build the DM-ELM model. The results indicate that combining hyperspectral information fusion with manifold learning dimension reduction algorithm can simultaneously distinguish the multiple defects of potatoes. [ABSTRACT FROM AUTHOR]

Remote sensing image classification is the main approach for rapidly obtaining regional land use/cover information and it has always been an important part in the field of remote sensing. How to improve the classification accuracy of remote sensing images is an urgent problem to be solved in remote sensing research. In traditional classification, only the spectral features of remote sensing image are used, while the texture and other features are ignored. Therefore, it is very common to see the object confusion in the classification result. In this paper, we took the Shijiazhuang Landsat 8 OLI remote sensing image data as the research area, and systematically studied object-oriented classification based on the spatial texture features of remote sensing images. Firstly, Gray Level Co-occurrence Matrix (GLCM) texture features and Gist texture features based on Gabor filter were compared and analyzed. The average J-M distance method was used to evaluate the sample separability and to choose optimal texture features of GLCM. Subsequently, the optimum index factor (OIF) was applied to obtain the best combination of the two texture features. Secondly, the segmentation scale of object-oriented classification was studied in detail, meanwhile, the concept of “the optimal overall segmentation scale” was proposed, which was based on the ratio between maximum area and the number of the objects in classification result. Finally, two object-oriented classification methods, K-Nearest Neighbor (KNN) method and Support Vector Machine (SVM) method, were used to classify the texture data and the original data, and the accuracy of assessment results were compared using three traditional supervised classification methods. The results indicated that the fusion of texture features could improve the accuracy of classification to some extent. The overall classification accuracy based on texture data using object-oriented SVM and object-oriented KNN increased by 3.67 and 3.33 percentage points, respectively, compared with the results based on original data. Object-oriented SVM method based on texture data had the highest classification accuracy with overall classification accuracy of 85.67%, and with Kappa index of 0.81. Although the classification accuracy of the texture-based supervised classification was improved compared with the supervised classification based on original data, the accuracy was far lower than the value with object-oriented method. For original data, the overall classification accuracy of object-oriented KNN increased by 4.33%, 3.99% and 2.00%, respectively, compared with Maximum Likelihood Classification (MLC), Mahalanobis Distance Classification (MDC) and SVM method. The overall classification accuracy of object-oriented SVM increased by 8.33%, 7.99% and 6.00%, respectively, compared with three supervised methods. After fusing Gist texture features, the overall classification accuracy of object-oriented KNN had increased by 6.33%, 4.66% and 2.66% respectively compared with MLC, MDC and SVM. Whereas the overall classification accuracy of object-oriented SVM increased by 10.67%, 9% and 7.00%, respectively, compared with three supervised methods. Object-oriented SVM method is more sensitive to texture features with the maximum increase of classification accuracy. In future study, more texture features need to be considered to extend the application range of remote sensing classification. In summary, the texture feature has positive effect on improving the accuracy of remote sensing classification, and the application of Gist textures have great potential in object-oriented classification. Moreover, it can also be found that object-oriented method is suitable for classifying medium resolution remote sensing image. The research method in this paper not only gives a valuable reference for other kinds of remote sensing images, but also provides an effective approach for the extraction of regional land use/cover information. [ABSTRACT FROM AUTHOR]

Due to high protein, low fat, high vitamin and mineral content, beef is regarded as an important meat item which is consumed by human in regular basis. In addition, with the improvement of people's living standard, different quality parameters such as beef tenderness, water retention and other edible quality indicators have become important factors to consumers' satisfaction. Sensory evaluation and shear force measurement are primarily two methods for measurement of beef tenderness, which are now considered as traditional methods. However, with improvement in technology, optical methods are playing important role for rapid, real-time, non-destructive and accurate detection of meat quality parameters. Near-infrared method and hyperspectral imaging method are important methods for non-destructive measurement of beef tenderness. This study focused on the application of hyperspectral image analysis for non-destructive detection of beef tenderness at different points of the sample. Considering the measurement complexity due to complex beef structure as well as the disadvantage of near infrared spectrum for single point detection, this study made use of hyperspectral imaging system to obtain three-dimensional information from each sample. Fifty-six beef samples were obtained from the slaughter house Yuxiangyuan in Beijing and were cut into 3 cm thick slices, vacuum packaged and then stored in 4°C environment before hyperspectral image acquisition. Reflectance spectrum was obtained for each sample from acquired images. Stepwise regression and genetic algorithm (GA) were used to identify the feature wavelengths, thus reducing the high dimensionality of the hyperspectral data. Eight feature wavelengths were selected, which were used to develop prediction model for measuring the shear force value at each point in the sample. By the principal component analysis (PCA), every sample could acquire 5 PCA images. Gray level co-occurrence matrix (GLCM) was applied to extract 8 main texture parameters from the characteristic band image and the 3 PCA images. Eight wavelengths were extracted from these 8 texture characteristic parameters respectively, with a total of 64 texture variables. Also, from five PCA images 40 texture variables were extracted. These parameters were used to establish a model based on linear discrimination analysis (LDA) and support vector machine (SVM) for beef tenderness. Because the number of principal components has a greater impact on the accuracy of discriminant model, before the training of SVM model, it first needed to select the appropriate number of principal components. Finally, three PCA images which included 95.37% information of every sample were chosen to extract texture parameters. With texture parameters of band images, the prediction set's accuracy of the discriminant model was 88.89%. There had some shortcomings that on one hand, the band images had great influence on the result, and on the other hand, some information may be missed with selected band images, so the texture features of the image of sample main component were used to establish classification model. After analysis and comparison, the accuracy of main component model was 94.44%, which was higher than that of band image model. In addition, the LDA method was more applied to establish classification model than SVM method. The results show that the image texture features can determine the distribution of beef tenderness. The results can provide the theoretical reference for the research on rapid non-destructive detection technique of beef tenderness. [ABSTRACT FROM AUTHOR]