Object detection and tracking method based on spiking neuron network

Aiming at the problems of large parameter capacity and heavy dependence on acceleration framework for complex target detection and tracking model, the conversion method between artificial neural network and impulse neural network is studied in this work to implement target detection and tracking task based on the spiking neural network. Firstly, the convolution modules Alexnet, vgg11-5c and vgg16-8c of the full convolution twin network Siamese-FC are properly adjusted and cut to adapt with the characteristics and limitations of impulse neurons, and then the Norm-SiamFC network model is constructed and trained. Finally, the migration parameters are adopted to construct a spiking network Spiking-SiamFC for object detection and tracking, which achieves good tracking effect and occupies less resources, thus the model can be widely utilized in edge devices with limited resource storage capacity.