Adjusting Xception Neural Network by Improved Human Evolutionary Optimizer for Fine Art Sorting

This study tackles the problem of fine-art image classification that suffers from both the softness of styles and the absence of labelled datasets. The state-of-the-art with CNNs leverages transfer learning and fine-tuning to overcome the scarcity of traditional datasets, which is not sufficient to extract the nuances within various fine art images. In this paper, a new approach is presented based on the combination of the Xception Neural Network and an Improved version of Human Evolutionary Optimizer (IHEO) to overcome these limitations. The model architecture of the Xception model, which uses a depth-wise separable convolution, allowing for a better capture of complex features, and introduced specialized IHEO, to enhance the task performance on fine art classification tasks. Through extensive experiments on various fine art datasets, such as WikiArt, WGA, and TICC, they achieved better accuracy and robustness over the state-of-the-art methods. These results demonstrate the promise of the Xception/IHEO approach in analyzing images in computational arts and classifications. There is also an analysis of deep representation transferability, where fine-tuning Xception on sentiment forecasting and scene identification is shown to outdo models pre-trained on object recognition. The challenges of fine art classification discussed herein in conjunction with methods for domain-specific weight initialization and optimization will allow for significantly more accurate and interpretable systems in the domain of fine art analysis going forward.