Efficient Processing Of Convolutional Neural Networks On The Edge: A Hybrid Approach Using Hardware Acceleration And Dual-Teacher Compression