My validation loss is fluctuating in between 0.6 and 1.2 and validation accuracy is between 60 and 65%

Question

Im training Mobilenet model with 531 images and using 6 types of data augumentationTechniques.Training accuracy is improving to 90 and loss is decreasing.But model is not generalizing well using validation set.

Usage:Adam(0.001) optimizer,batch size=32,binary crossentropy,two classes(normal and pathological),sigmoid output activation function.L2 regularizer(0.001),also using dropout layers(0.2).What could be the reasons and solutions.