Image Analysis for Agricultural Products and Processes

Variability of weed infestation needs to be assessed for site-specific weed management. Since manual weed sampling is too time consuming for prctical applications, a system for automatic weed sampling was developed. The system uses bispectral images, which are processed to derive shape features of the plants. The shape features are used for the discrimination of weed and crop species by using a classification step. In this paper we evaluate different classification algorithms with main focus on k-nearest neighbours, decision tree learning and Support Vector Machine classifiers. Data mining techniques were applied to select an optimal subset of the shape features, which then were used for the classification. Since the classification is a crucial step for the weed detection, three different classification algorithms are tested and their influence on the results is assessed. The plant shape varies between different species and also within one species at different growth stages. The training of the classifiers is run by using prototype information which is selected manually from the images. Performance measures for classification accuracy are evaluated by using cross validation techniques and by comparing the results with manually assessed weed infestation.