Digital Image Vegetation Analysis with Machine Learning

We propose computer vision based approach for effectively computing the vegetation coverage of the image to determine the structure of the vegetation and to understand wildlife habitat. To deal with the variation of lighting condition, two-stage segmentation strategy is applied. Firstly, texture information is used to roughly classify the vegetation and the reference blackboard at each position using Support Vector Machine. And then a K-means based adaptive color model is used to refine the segmentation result in pixel level. We evaluate our approach on our dataset, and the results demonstrate that the proposed method is robust to environment changing, and color instability. For blackboard localization, we tested 200 images and the accuracy is approximately 93%. For grass detection and coverage computation, the error rate is approximately 3%.