An instance-based learning approach for thresholding in crop images under different outdoor conditions

作者：Arroyo Javier^*; Guijarro Maria^*; Pajares Gonzalo

来源：Computers and Electronics in Agriculture, 2016, 127: 669-679.

DOI：10.1016/j.compag.2016.07.018

• Image segmentation
• Symbolic data analysis
• Match distance
• Agricultural images
• Thresholding
• Binarization

We propose an instance-based learning approach for segmentation of crop images. The method proposed is able to automatically discriminate the green textures (crop and weeds) from the rest of the ground under different outdoor conditions, namely light conditions and stages of crop growth. For this purpose a set of images that reflects all the possible conditions to be faced is required and each images should have assigned its optimum threshold by an expert.
The instance-based method proposed is a kappa-Nearest Neighbors (kappa-NN) algorithm. Our kappa-NN is borrowed from the field of symbolic data analysis that is a paradigm for data representation where instances can be described by variables that account for observed variability, such as, in our case histogram variables.
Our method is compared with the well-known automatic thresholding methods, such as the mean value and Otsu's. The method proposed provides good results for all the different conditions analyzed, including burned and saturated images.

• 出版日期2016-9