Computations adapted from the interactions of neurons in the nervous system may be a capable platform that can create powerful machines in terms of cognitive abilities such as real-time learning, decision-making and generalization. In this regard, here an intelligent machine based on the basic and approved mechanisms of the nervous system has been proposed. Therefore, the input layer of the presented machine is adapted from the retinal model and the middle layer and the output layer is composed of population of pyramidal neurons/ interneurons, AMPA/GABA receptors, and excitatory/inhibitory neurotransmitters. A machine that has a bio-adapted structure requires a learning based on biological evidence. Similarly, a new learning mechanism based on unsupervised (Power-STDP) and reinforcement learning procedure (Actor-Critic algorithm) was proposed which was called PSAC learning algorithm. Three challenging datasets MNIST, EMNIST, and CIFAR10 were used to confirm the performance of the proposed learning algorithm compared to deep and spiking networks, and respectively accuracies of 97.7%, 97.95% (digits) and 93.73% (letters), and 93.6% have been obtained, which shows an improvement in accuracy compared to previous spiking networks. In addition to being more accurate than the previous spike-based learning methods, the proposed learning approach shows a higher convergence speed in the training process. Although the obtained classification accuracies are slightly lower than deep networks, but higher training speed, low power consumption if implemented on neuromorphic platforms, and unsupervised learning are the advantages of the proposed network.