Fingerprint recognition is a vital biometric technology employed in various security systems due to its high reliability and uniqueness. This paper presents a novel FINGPRA-NET for fingerprint recognition system based on deep learning. The proposed fingerprint recognition algorithms that incorporate feature extraction, pattern matching, and deep learning techniques to enhance recognition accuracy and computational efficacy. Initially, the gathered images are denoised using Laplacian filter to highlight regions of fast intensity variation in an image, which is useful for detecting the edges. The atrous spatial pyramid pooling (ASSP) layer is integrated in deep learning based modular neural network to enhance the performance and generalization capabilities of the model. This combination leverages the strengths of both ASPP (for capturing context at multiple scales) and MNN (for specialized detection tasks), resulting in improving the accuracy in fingerprint recognition. Coyote Optimization Algorithm (COA) is used to find the best correspondences between minutiae points in different fingerprint images. The experimental fallouts prove that the proposed FINGPRA-NET method achieves an overall accuracy of 98.92% outperforming traditional methods. The high accuracy and low error rates signify the model's effectiveness in distinguishing between authentic and non-authentic fingerprints.