When training your own model, the accuracy rate is 0.98, and the loss is finally stable at 0.78

I built a model, its accuracy and high, reaching 0.98, but the value of loss finally stabilized at around 0.78, not falling, It is reasonable to say that the accuracy and loss reflect the difference between predicted value and the real value,they should be inversely proportional.If the accuracy is high, the loss should be low.
The following figure, DerNet, is a model I built. Compared with VGG and Resnet, the other parameters are the same. I want to know what is the factor that affects the loss.