Nowadays, Deep neural networks (DNNs) including QNN (Quantised Neural Network) and BNN (Binary Neural Network) are widely used in wide range of artificial intelligence (AI) applications including image processing, object detection and robotics. However, implementing these deep neural networks on embedded devices is a major bottleneck due to the massive requirement of computation and storage. Therefore, use of techniques which can carry out efficient processing of DNNs in AI systems and improve inference time and execution time keeping into consideration the critical performance parameters such as accuracy or hardware cost is very important. In this paper, an efficient implementation and analysis of QNN- and BNN-based pattern recognition techniques have been done. The FPGA PYNQ-Z2 has been used for QNN and BNN algorithms and has shown improved performance in terms of inference time, efficiency and hardware. The comparative analysis of implementation of BNN and QNN has been carried out on the basis of accuracy, weight bit error, RoC curve, and execution speed. The proposed Image detection algorithms has produced significant results on identifying images with a percentage decrease in inference time of about 12×, 7× increase in the accuracy, 95% decrease in execution time in comparison to the implementation on software for both the above stated parameters. The plotted Roc Curve indicated good performance levels.