Synthesizing feature agents using evolutionary computation
Introduction
Image segmentation and labeling of terrain regions is an important task in remote sensing application (Mvogo et al., 2000; Jeon et al., 2002; Katartzis et al., 2001; Segl and Kaufmann, 2001; Acqua and Gamba, 2001; Dong et al., 2001). The quality of this task is primarily dependent on the data and features used as input. There are many kinds of features that can be used, and the question is what are the appropriate features or how to synthesize features, particularly useful for segmentation/labeling, from the primitive features extracted from images. Usually, there are almost infinite ways of combining primitive features to form synthesized composite ones. It is the human image experts who, relying on their rich experience, figure out a smart way of combination. The task of finding a good combination is equivalent to finding a good point in the search space of agents formed by the combination of primitive operations (also called primitive operators) on images.
However, limited by their speed, previous experience or even bias, the human experts can only try a very limited number of conventional combinations. Genetic programming (GP), on the other hand, may try many unconventional combinations that may yield exceptionally good results. Also, genetic programming can explore a much larger portion of agent space due to the inherent parallelism of GP and the speed of computer. The search performed by GP is guided by the goodness of agents in the population. As the search proceeds, GP will gradually shift the population to the portion of the space that contains good agents. The crossover operation is the major mechanism used by GP to search the agent space and the mutation operation is employed to increase the diversity of the population to avoid the premature convergence. However, in the traditional crossover and mutation, their locations are randomly selected, leading to disrupting the effective components (subtree in this paper) within agents and thus greatly reducing the efficiency of GP. It is very important for GP to identify and keep those effective components.
In this paper, we use genetic programming to generate feature agents for terrain labeling. The individuals in our GP-based learning are agents represented by binary trees whose internal nodes represent the pre-specified primitive operators and the leaf nodes represent the original image or the primitive feature images generated from the original image. The primitive feature images are pre-determined, and they are not the output of the pre-specified primitive operators. After applying an agent on the original image or the primitive feature images, the output image of the agent is segmented to yield a binary image or mask. The binary mask is used to extract the particular kind of terrain regions from the original image. To improve the efficiency of GP, we propose smart crossover and smart mutation to identify and keep the effective components of an agent.
Section snippets
Motivation for intelligent GP operators
Search operators are one of the most important parts of any machine learning system. They define the manner in which the learning system moves through the space of candidate solutions. In this paper, crossover and mutation are major search operators used by GP to search the huge agent space. Finding good agents is a tough task and it is very important to design intelligent crossover and mutation operators in order to improve the efficiency of GP.
As it is well known, crossover is the predominant
Technical approach
In our GP-based approach, individuals are agents represented by binary trees. The search space of GP is the space of all possible feature agents. The space is very large. To illustrate this, consider only a special kind of binary tree, where each tree has exactly 30 internal nodes and one leaf node and each internal node has only one child. For 17 primitive operators and only one primitive feature image, the total number of such trees is 1730. It is extremely difficult to find good agents from
Experiments
Various experiments were performed to test the efficacy of genetic programming in extracting regions of interest from real synthetic aperture radar (SAR) images. We show three selected examples. In each of three examples, GP is applied to the training images to generate an agent for a particular kind of terrain region and the agent is then applied to the testing images. It is worth noting that the training and testing images are different images and the ground truth is used only during
Conclusions
In this paper, we presented a basic approach that uses genetic programming to evolve agents for extracting terrain regions in remotely sensed images. In order to improve the efficiency of genetic programming, we proposed smart crossover and smart mutation that can identify and keep the effective ROI extraction components in agents. We used SAR imagery to demonstrate the approach for extracting terrain regions. Our experimental results showed that GP can find good agents to effectively extract
Acknowledgements
This research is supported by AFRL grant F33615-99-C-1440 and NSF grant IIS-0114036. The contents of the information do not necessarily reflect the position or policy of the US Government.
References (19)
- et al.
Target detection in SAR imagery by genetic programming
Adv. Engrg. Software
(1999) - et al.
Detection of urban structures in SAR images by robust fuzzy clustering algorithms: The example of street tracking
IEEE Trans. Geosci. Remote Sensing
(2001) - et al.
Genetic Programming––An Introduction on the Automatic Evolution
(1998) - Bhanu, B., Lin, Y., 2002. Learning composite operators for object detection. In: Proc. Genetic and Evolutionary...
- D'haeseleer, P., 1994. Context preserving crossover in genetic programming. In: Proc. IEEE World Congress on...
- et al.
Segmentation and classification of vegetated areas using polarimetric SAR image data
IEEE Trans. Geosci. Remote Sensing
(2001) - et al.
Digital Image Processing
(1993) - Ito, T., Iba, H., Sato, S., 1998. Depth-dependent crossover for genetic programming. In: Proc. IEEE Internat. Conf. on...
- et al.
Road detection in spaceborne SAR images using a genetic algorithm
IEEE Trans. Geosci. Remote Sensing
(2002)
Cited by (3)
Vision and evolution: State of the art
2016, Natural Computing SeriesFeature fusion of mechanical faults based on evolutionary computation
2007, Insight: Non-Destructive Testing and Condition MonitoringEvolutionary computer vision
2007, Proceedings of GECCO 2007: Genetic and Evolutionary Computation Conference, Companion Material