Variable-Length Deep Convolutional Neural Networks by Internet Protocol Chimp Optimization Algorithm for Underwater Micro-Target Classification

Due to the complexities of classifying acoustic targets, the use of conventional and conventional methods has been seriously challenged. Deep-convolutional neural networks (DCNNs), on the other hand, are among the safest methods for solving image problems. However, designing a DCNN architecture for a subject by raising the dimensions of underwater targets can be very challenging. To solve this problem, this paper uses the Chimpanzee Optimization Algorithm (ChOA) to find the best architecture for DCNNs. In this regard, three innovations based on the standard ChOA are proposed in order to achieve an audio classifier with minimal complexity and high accuracy. First, a unique Internet Protocol (IP) address-based encoding method is developed that makes it easier to encode DCNN layers for chimpanzee vectors. Then, an attenuation layer is recommended to achieve DCNNs of different lengths, some of which cover the chimpanzee vector. As a third innovation, the learning process of the big data set is divided into smaller parts, which are then randomly evaluated. Then, after collecting the required data and performing a comprehensive simulation for architecture, the evaluation error is 0.000827, which is obtained in a total time of 1012. The results show that the proposed method, in addition to increasing the accuracy of model training, significantly saves computational time.