A Two-Stage Approach Using Genetic Algorithm and Genetic Programming for Remote Sensing Crop Classification

Crop classification is an important task in remote sensing image analysis. To effectively classify crops, it is necessary to extract or obtain a set of effective features from raw pixels. However, existing methods have several limitations, including poor interpretability of the learned models and the requirements of sufficient training data and domain expertise. To address this, this paper develops a two-stage approach using Genetic Algorithm (GA) and Genetic Programming (GP) to automatically learn a feature set that can effectively classify crops using remote sensing images. In the first stage of the new approach, a GP method is applied to automatically construct a set of high-level features by evolving tree-based solutions. In the second stage, a GA method is employed to select a small subset of features from the constructed features and the original features by removing redundant ones. The performance of the new approach is evaluated on three datasets in two scenarios, i.e., classifying four main crop types and all crop types, respectively. The results demonstrate that the new approach achieves more accurate crop classification compared with eight competitive methods.