The validation images with the distribution of few contaminated kernels mixing among the healthy were classified using the trained classifier and parameters in learning images. Classification results are shown as in Fig. 4 and Table 2. The pixel-wise classification maps in Fig. 4 indicate satisfactory result with most pixels classified correctly except for some mixed pixels at kernel margins. Table 2 gives the pixel-wise classification accuracies of 96.32% for breed A, 94.20% for B and 97.51% for C. Although there exists some misclassified pixels, reconsidering the quality of each peanut at a kernel-scale spatially can smooth them and give reliable decision of whether a kernel is fungi-contaminated. More importantly, kernel-scale identification maps can clearly tell which peanut is contaminated and should be separated.