Progress and development of any society depends on having accurate information from its environment. This information extraction can be achieved in several ways. One of the ways you can manually extract the position of the objects from satellite images that very time consuming. Automatic extraction of the features from satellite imagery has provided sufficient accuracy can be considered as an appropriate method to replace manual methods. This paper addresses Adaboost training algorithm and its sensitivity analysis by using Haar-like features. As the algorithm’s name imply, a set of images are required to train the algorithm and to build positive and negative training schemas. The images are extracted form three features such as automobile, airplane and oil reservoirs; moreover the images are utilized for the training of the algorithm. After the algorithm’s training for each aforementioned features by using enough quantity of schemas, the training procedure is completed and ready for the feature extraction from the satellite images. This research have studied on WorldView-3 satellite images with the spatial resolution of 40 centimeters. In present study and after the feature extraction, the algorithm’s results are compared to results which is derived from the manual extraction. The comparison shows the high efficiency and precision of the algorithm. In the automobile and oil reservoirs accuracy and completeness algorithms able to reach 90percent. The main characteristic of this algorithm is to run quickly in very large high resulation satellite imagery. In order to analyze the sensitivity of the, the interdependency of the algorithm’s precision and completeness is studied in comparison with a verity of parameters of algoritm’s including the number of training stage’s and completeness of each stage so on that can be very expediting and helpful in the training of the algorithm for the extraction of other features.