Study On Intelligent Understanding Algorithm Of Inspecting Video Based On Depth Learning Technology

In recent years,along with the rapid development of power grid construction and the enlargement of the scale of power lines,the secure operation of the power grid is facing a new challenge.And also,new,efficient and smart management means and methods are urgently needed.Traditional power inspection relies on man-power.People need to climb to tower and travel across mountains.And there are many problems such as difficult operation,high risks and low efficiency.In recent years,drone technology and image recognition technology have made considerable progress,bringing new tools and ideas to power line inspections.And relying on drones and image recognition technology can greatly improve inspection efficiency.It is also a hot research direction at present.Deep learning is currently widely used in the fields of voice,image and text,and has achieved great benefits in industrial applications.The author introduces deep learning technology into the understanding of patrol video images to help people find problems and improve efficiency.The author selected two different deep learning image recognition frameworks based on “classification” and “regression”.They are used to identify transmission towers,trees,insulators,bird nests,and other objects in video.The author prepared the pictures of these four objects as the original data to train the model,and finally observed and compared the performance differences between the two models.The author has obtained satisfactory recognition accuracy in the limited training samples,and confirmed the feasibility of deep learning for inspection image identification.The author gives the principle of frame selection under different demand scenarios,chooses "YOLO" under the circumstance of speed preference,and selects "Faster R-CNN" in the scenario of more concerned accuracy.