Optimal color space selection method for plant/soil segmentation in agriculture

Display Omitted A new training algorithm is proposed for agricultural color classification problems.The proposed method selects the optimal color space and channels for each scenario.Applied to estimate accurately and efficiently the percentage of green cover (PGC).Developed an application called ACPS (Automatic Classification of Plants and Soil). Color analysis techniques in agriculture should be able to deal with non-trivial capture conditions such as shadows, noise, pixel saturation, low lighting, different varieties of crops and intrinsic parameters of the cameras. Previous studies have shown the importance of selecting the optimum color space for each application domain. This paper presents a new probabilistic approach to color processing capable not only to create optimum color models for the plant/soil segmentation, but also to select the most adequate color space for each problem. The system evaluates all the possible alternatives, producing color models in the optimum space and channels. Thereby, the dependences on the kind of crop, camera and capture conditions are avoided, since the method is adapted to the training conditions. The basis of the proposal is the use of non-parametric models for the probability density functions of plant/soil colors. The proposed method has been implemented and validated in a new software tool, called ACPS (Automatic Classification of Plants and Soil), thus proving its practical feasibility. The final purpose of this system is the analysis of the vegetal ground cover, in order to obtain the PGC (percentage of green cover) parameter. The ACPS software has been developed to be used by professionals, researchers, technicians and anyone working in the agricultural area. Furthermore, the models created can be exported to a defined file format which can be used in applications in the cloud, mobile devices and compact controllers that are currently being developed.