Image recognition based on THGS algorithm to optimize ResNet-18 model

Abstract

In order to quickly and accurately identify brown spot images, an improved THGS-ResNet-18 recognition model is proposed in this paper. Firstly, the Hunger Game search algorithm is improved by using Tent chaos mapping to solve the problem of excessive randomness in the population initialization of the Hunger Game search algorithm. Secondly, the hyperparameters of the improved Hunger Game search algorithm are optimized for the ResNet-18 model. Finally, the improved model THGS-ResNet-18 is applied to identify 5064 rice leaf images, and compared with four other ResNet-18 models improved by swarm intelligence algorithm on seven evaluation indicators. Experiments show that the model proposed in this paper has improved accuracy 5.22%−6.09%, sensitivity 3.53%−5.31%, specificity 7.38%, precision 6.95%−7.13%, recall 3.53%−5.31%, f-measure 5.22%−6.20%and g-mean 5.24%−6.13%.