| abstract |
The invention discloses a smoke identification method suitable for forest fires, comprising the steps of: 1. building an image database; 2. training and testing SVM; 3. building an automatic smoke monitoring platform; 4. image collection and background frame image modeling; 5. Background difference processing; 6. Logical operation to obtain moving foreground targets; 7. Smoke identification of forest fires. The present invention collects common cloud and fog, morning glow, sunset, and water vapor pictures that are easy to confuse, and conducts SVM training with less data requirements, selects textures redundantly to ensure high smoke recognition accuracy, and uses continuous multi-frame images and background frame images through continuous multi-frame images and background frame images. Perform differential operation to effectively extract tiny moving targets, and then perform logical OR operation on continuous multi-frame real-time images of forest areas to obtain all current moving target information, and eliminate random interference by performing logical AND operation on multiple ORed images. The moving target can obtain an image that retains the moving foreground target, and the smoke recognition effect is good. |