Complex chemical processes occur in during cocoa bean fermentation. To select well-fermented beans, experts take a sample of beans, cut them in half and visually check its color. Often farmers mix high and low quality beans therefore, chocolate properties are difficult to control. In this paper, we explore how close-range hyper- spectral (HS) data can be used to characterize the fermentation process of two types of cocoa beans (CCN51 and National). Our aim is to find spectral differences to allow bean classification. The main issue is to extract reliable spectral data as openings resulting from the loss of water during fermentation, can cover up to 40% of the bean surface. We exploit HS pan-sharpening techniques to increase the spatial resolution of HS images and filter out uneven surface regions. In particular, the guided filter PCA approach which has proved suitable to use high-resolution RGB data as guide image. Our preliminary results show that this pre-processing step improves the separability of classes corresponding to each fermentation stage compared to using the average spectrum of the bean surface.
In this paper, we propose a novel semi-supervised graph leaning method to fuse spectral (of original hyperspectral (HS) image) and spatial (from morphological features) information for classification of HS image. In our proposed semi-supervised graph, samples are connected according to either label information (labeled samples) or their k-nearest spectral and spatial neighbors (unlabeled samples). Furthermore, we link the unlabeled sample with all labeled samples in one class which is the closest to this unlabeled sample in both spectral and spatial feature spaces. Thus, the connected samples have similar characteristics on both spectral and spatial domains, and have high possibilities to belong to the same class. By exploiting the fused semi-supervised graph, we then get transformation matrices to project high-dimensional HS image and morphological features to their lower dimensional subspaces. The final classification map is obtained by concentrating the lower-dimensional features together as an input of SVM classifier. Experimental results on a real hyperspectral data demonstrate the efficiency of our proposed semi-supervised fusion method. Compared to the methods using unsupervised fusion or supervised fusion, the proposed semi-supervised fusion method enables improved performances on classification. Moreover, the classification performances keep stable even when a small number of labeled training samples is available.
Spectral unmixing aims at finding the spectrally pure constituent materials (also called endmembers) and their respective fractional abundances in each pixel of the hyperspectral image scene. One important issue in hyperspectral data unmixing is the initialization of endmembers. Most unmixing methods initialize their endmembers by randomly selecting a specified number of pixels from the data or by vertex component analysis, which limits their performance in practice. We propose an endmember initialization method for hyperspectral data unmixing. Our initial endmembers include some of the true endmembers, which improves the accuracy of hyperpspectral unmixing effectively. The experimental results on both synthetic and real hyperspectral data illustrate the superiority of the proposed method compared with other state-of-the-art approaches.