This paper presents a planar diffractive neural network (pla-NN) for direct electromagnetic (EM) information processing. The pla-NN is designed to perform direct signal processing in the microwave frequency range, enabling efficient and flexible object recognition in EM space. The architecture is based on printed circuit fabrication, allowing for high integration, conformal design, and flexible extension. The network consists of propagation and modulation layers, with the modulation layers containing phase shifters that enable trainable unitary transformations. The pla-NN is trained using gradient-based algorithms to optimize phase distributions for desired signal mapping. The network is validated on the fashion-MNIST dataset, demonstrating high classification accuracy. The pla-NN is also tested with different data injection methods, including direct injection, Fourier space injection, and propagation mingled injection, achieving classification accuracies of 90.18%, 88.99%, and 89.60%, respectively. The system is further tested with real-world applications, including object recognition in EM space, where it successfully identifies four standard 3D objects with high accuracy. The pla-NN is shown to have low latency, with an estimated processing latency of 2.3 ns. The network is also scalable, with the ability to be extended through layer stacking and flexible conformal designs. The pla-NN is compared to other EM neural network architectures, such as deep diffractive NN (D²NN) and optical neural network (ONN), and is found to be more suitable for lower-frequency applications due to its use of microstrip lines instead of optical waveguides. The pla-NN is also energy-efficient and has potential applications in high-performance computing, wireless sensing, and flexible wearable electronics. The paper concludes that the pla-NN represents a new direction in microwave signal processing and offers an alternative architecture for EM-based neural computing.This paper presents a planar diffractive neural network (pla-NN) for direct electromagnetic (EM) information processing. The pla-NN is designed to perform direct signal processing in the microwave frequency range, enabling efficient and flexible object recognition in EM space. The architecture is based on printed circuit fabrication, allowing for high integration, conformal design, and flexible extension. The network consists of propagation and modulation layers, with the modulation layers containing phase shifters that enable trainable unitary transformations. The pla-NN is trained using gradient-based algorithms to optimize phase distributions for desired signal mapping. The network is validated on the fashion-MNIST dataset, demonstrating high classification accuracy. The pla-NN is also tested with different data injection methods, including direct injection, Fourier space injection, and propagation mingled injection, achieving classification accuracies of 90.18%, 88.99%, and 89.60%, respectively. The system is further tested with real-world applications, including object recognition in EM space, where it successfully identifies four standard 3D objects with high accuracy. The pla-NN is shown to have low latency, with an estimated processing latency of 2.3 ns. The network is also scalable, with the ability to be extended through layer stacking and flexible conformal designs. The pla-NN is compared to other EM neural network architectures, such as deep diffractive NN (D²NN) and optical neural network (ONN), and is found to be more suitable for lower-frequency applications due to its use of microstrip lines instead of optical waveguides. The pla-NN is also energy-efficient and has potential applications in high-performance computing, wireless sensing, and flexible wearable electronics. The paper concludes that the pla-NN represents a new direction in microwave signal processing and offers an alternative architecture for EM-based neural computing.