Novel modified convolutional neural network and FFA algorithm for fish species classification

Biologists and marine ecologists' vital task is to classify fish species regularly to assess the relative profusion of fish species in their native environments and track population fluctuations. Traditional fish species classification procedures were time-consuming, labor-intensive, and costly. Consequently, an automated method is required, although improving its effectiveness remains a challenge. This research proposed a Modified Convolutional Neural Network (MCNN)-based on different classifications of fish species to augment the proposed methodology’s performance. Pre-processing, segmentation, feature extraction, feature optimization, and classification are the five phases in the proposed scheme. The input image is enhanced by using Thinning operation with Hit Miss function in the preprocessing stage. From the image, fish is segmented using Region of Interest and Hadamard Control Firefly Algorithm (HCFA). By using HCFA, the boundary is selected and the opted boundary is extracted using a canny edge detector. The Speeded Up Robust Feature feature is extracted from the segmented part after the categorization of fish. By considering classification accuracy as the fitness function, the feature is optimized by wielding the same HCFA. MCNN classifies the different fish species by giving the obtained feature as input to it. The proposed system encompassed ‘5’ diverse species and a total of 2000 images of every fish kind along with were classified. Grounded on the performance metrics, the proposed methodology’s performance is analogized with the prevailing research methodologies. This system acquired 97.55% accuracy; thus, it is suited perfectly for the fish species classification.