Visual Significance Target Detection Algorithm under Weak Supervised Learning
Download as PDF
Objective: To improve the existing target detection algorithm under multi-example learning and to surpass the existing target detection algorithm under weak supervised learning. Methods: To obtain the underlying visual features of the significance region according to the existing significance methods and train the performance of the target features. The visual significance objective is studied by means of random vector to obtain the weight of representation in the significance. The ROC curve of the iterative significant graph was calculated to find the optimal solution and the optimal weight of the target. Results: The application of visual significance target detection algorithm greatly promoted the development of weak supervised target detection algorithm, and gradually replaced the traditional multi-example learning method. In particular, the accuracy of Pascal VOC 2007 was significantly improved after the joint algorithm was adopted, reaching 79.3%. Conclusion: The visual significance target detection algorithm under weak supervised learning simulates the process of human perception of specific attention objects in vision. The detection accuracy of this algorithm is better than many existing algorithms, and it can effectively detect the visual significance target.
Weak Supervision Learning; Visual Significance; Target Detection Algorithm; Multi-example Learning