In this paper, a convolutional neural network model is used to identify biometric features and design a classifier. The network architecture is used to extract the biometric features of the image, followed by a novel biometric image classification algorithm that is able to simultaneously optimise the structural content of the weight parameters, bias parameters and hyperparameters of the extreme learning machine. In the first stage, the extracted features of the biometric image are reduced in dimension using a pre-trained DenseNet-121 network, and in the second stage, the optimized ELM classifier is trained for prediction. In order to reduce the uncertainty of the ELM classifier in terms of random input weights and biases, chaotic initialisation, multiple swarm strategy and fuzzy logic optimisation flow algorithm are used, in addition to the search agent strategy, which is used to adjust the input weights and biases of the ELM. The CNN-based image recognition method and the optimized ELM classifier efficiently perform the identification and classification of biometric images. The CNN-based image recognition method is highly accurate at 97.26%, and its recognition efficiency is much higher than the other three models mentioned in the experiment. And the overall classification accuracy of the model based on the optimized classifier reaches 92.22%, which achieves a stable improvement in accuracy and effectively improves the recognition of biometric image classification.