Detection of dangerous water area during UAV autonomous landing

Aiming at the problem of water dangerous area detection faced by UAV during emergency autonomous landing, the features of water dangerous area are extracted from the image by neural network, the texture features of the image are obtained by HOG algorithm, and the features extracted by neural network and texture features are classified by support vector machine method (SVM). Then, the classifier is trained based on color features and regional texture features to detect the specific location of water hazard areas in the image. The experiment shows that the method has a good result in detecting the dangerous area of water during UAV autonomous landing, and the detection accuracy can reach more than 90%.