Abstract
Rapid advancement in remote sensing sensors has resulted in an enormous increase in the use of satellite imagery (SI) and images taken from unmanned aerial vehicles (UAVs) in a wide range of remote sensing applications. These applications include urban planning, environment monitoring, map updating, change detection, and precision agriculture. This paper focuses on an agricultural application of SI and UAV images. SI-UAV images possess high temporal, textural, and intensity differences due to rapid changes in agricultural crops with the passage of time. Feature points such as scale invariant feature transform (SIFT), oriented FAST and rotated BRIEF (ORB), and speeded-up robust features (SURF) are not invariant to such differences and underperform in SI-UAV image registration. To deal with this problem, we propose a new method that combines the strength of nearest neighbor (NN) and brute force (BF) descriptor matching strategies to register SI-UAV images. The proposed method is named NN-BF. For NN-BF first corresponding feature point descriptor matches are identified between SI-UAV images of the training set with overlap error. Then the corresponding descriptors are matched with the descriptors of SI images of the test set with NN strategy. The resulting descriptor matches are then further matched with the descriptors of UAV images of the test set using BF strategy. Finally, the descriptor matches obtained are processed with RANSAC to remove outliers and estimate a homography for image registration. Experiments are performed on an agricultural land image dataset. The experimental results show that the NN-BF method improves SIFT, SURF, and ORB feature point performances and also outperforms recently proposed feature matching strategies for remote sensing images. SIFT on average obtains 6.1% and 18.9% better precision scores than SURF and ORB with NN-BF, respectively. SIFT also obtains lower root mean square error than SURF and ORB with NN-BF.