| abstract |
The 3DCNN-based forest smoke detection and classification method for UAV images belongs to the field of computer vision, specifically involving deep learning, target detection and other technologies. The present invention first extracts the spatio-temporal information of multiple channels from the original frame sequence, and obtains the initial feature maps of multiple channels for the feature extraction of the convolution layer; Multi-scale feature extraction; finally connect the feature maps of each layer to obtain the feature vector, combine the SVM (Support Vector Machine) classifier, and complete the training and smoke detection and classification on the forest smoke video dataset. The present invention makes up for the shortcomings of difficult model design and large amount of calculation. The convolution layer and downsampling layer of different scales in the neural network can extract robust depth scene feature information, and can adapt to multi-angle and multi-scale shooting scenes; secondly, It has a good ability to describe the dynamic spatiotemporal characteristics of smoke. |