<p>Face sketch recognition involves matching hand-drawn or composite sketches to real facial photographs, and the current methods struggle with the visual gap between sketches and photos. Deep Learning (DL) models need large datasets and often fail with varied or poor-quality sketches. To overcome these limitations, the Archery Archimedes Optimization Algorithm-based Parallel Orthogonal Principal Network (AAOA_POPNet) is proposed for face sketch recognition. Here, the face generating module, initialization module and optimization module are considered. Initially, the input sketch image is subjected to latent space generation using the Deep Maxout Network (DMN) in the face-generating module. Then, the outcome and input sketch image are passed to the Scarcity-Generative Adversarial Network (GAN) in initialization module. Here, the latent space is updated by utilizing an optimization module, which consists of a face appearance module, and a Manifold Preservation Module. Then, the latent space from each module is optimized by utilizing AAOA. Here, AAOA is formed by hybridizing Archery Algorithm (AA) and the Archimedes Optimization Algorithm (AOA). Further, the real image is allowed for feature extraction, and the face sketch recognized output is obtained. At the same time, an input sketch image is allowed for facial landmark detection to attain the face sketch recognized output. Finally, both recognized outputs are allowed for face sketch recognition using POPNet, where the weights of POPNet are optimized using AAOA, and are formed by combining Parallel Orthogonal Deep Neural Network (PODNN) and Principal Component Analysis Network (PCANet). Additionally, AAOA_POPNet has attained 85.47% of accuracy, 0.842 of Matthews Correlation Coefficient (MCC), and 84.30% of precision.</p>