UAV Image Classification and Recognition Based on SVM and Non-Reference Adaptive Quality Evaluation

Image classification and recognition are the hotspots of UAV image processing and have broad application prospects. However it also has the disadvantage of heavy load and low efficiency to operators. In this paper, images are effectively processed as classification, layered and grade with a new type of algorithm combining with sparse matrix and SVM and feature extraction. Sparse matrix is used to classify images. And image quality features of wavelet transform, DCT, HAAR and entropy are extracted under image preprocessing to be normalized. Then, trainers and testers are constructed with cross-validated and multi-kernel parameters optimization under no reference. And the feedback of optionally retest result under full reference or semi-reference quality is modifying the trainer. Finally a new model with combing confused matrix is built to achieve adaptive quality evaluation.