A comparison of pixel-based and object-oriented image classification techniques for forest cover type determination in east Texas

• Main Author: Raines, Jason
• Format: Dissertation
• Language: English
• Subjects: agriculture, forestry and wildlife; remote sensing
• ISBN: 9780549755241; 0549755241

The ability to qualify and quantify the forest resources of east Texas through remote sensing techniques is a timely, cost-efficient manner to assist resource managers and policy-makers in planning and decision-making. Improved image classification techniques would reduce costs and timelines of resource inventory and analysis as well as increase the accuracy of cover type maps produced from medium resolution satellite imagery. Object-oriented image classification techniques take advantage of the spatial autocorrelation innately present in remotely sensed data, while pixel-based classifiers focus solely on the spectral signature of each pixel. Leaf-off Landsat TM data were used to classify land cover in east Texas into four classes. Object-oriented classification techniques were compared with traditional pixel-based classification methods. Accuracy assessment was conducted to generate error matrices for each method tested. Statistical tests showed supervised object-oriented classification using Feature Analyst(TM) performed significantly poorer than all other methods tested except unsupervised pixel-based classification.