This research aims to identify an optimum classification engine and the parameters that maximize the accuracy of in-field machine vision classification of tree species based on their bark color and texture, in conditions of large intraspecies color and texture variations. Feature vectors based on hue and saturation histograms extracted from images for color classification, and feature vectors based on the histogram of gradients (HOG) and the histogram of binary patterns (HBP) algorithms extracted for texture classification were fed to the nearest neighbor (NN) and the support vector machine (SVM) classification engines for assessment. Libraries for multiple, simultaneous, parallel image pipelines for training and classification were created for each classification engine and were used for sensitive analysis of parameters such as classification method, distance metrics, k-value, grid size, or number of histogram bins. For the relatively small image dataset used in this study, the best classification scores were obtained for both color (75.4% accuracy) and texture (63.7% accuracy) using an SVM with a grid size of 1 X 1 and 180 histogram bins. The classification score was further improved to 84.4% accuracy using a late color-texture fusion of the best two classifiers. The paper describes the classification methods used in the study, the metrics used for assessing the classification results, and the sensitivity analysis used to optimize the classification method. The classification libraries developed in this study will be deployed on an autonomous rover for automated woodland mapping and forest farming.