In the paper we introduce a novel bipolar morphological neuron and bipolar morphological layer models. The modelsuse only such operations as addition, subtraction and maximum inside the neuron and exponent and logarithm as activationfunctions for the layer. The proposed models unlike previously introduced morphological neural networks approximatethe classical computations and show better recognition results. We also propose layer-by-layer approach to train thebipolar morphological networks, which can be further developed to an incremental approach for separate neurons to gethigher accuracy. Both these approaches do not require special training algorithms and can use a variety of gradient descentmethods. To demonstrate efficiency of the proposed model we consider classical convolutional neural networks and convertthe pre-trained convolutional layers to the bipolar morphological layers. Seeing that the experiments on recognition ofMNIST and MRZ symbols show only moderate decrease of accuracy after conversion and training, bipolar neuron modelcan provide faster inference and be very useful in mobile and embedded systems.
展开▼