Enhancing Image Recognition Using Hierarchical Neural Networks Tree in Computer Vision

Image classification is an important research direction in the field of artificial intelligence and machine learning. However, it faces substantial challenges when dealing with visually similar items that possess different characteristics. To address this issue, this article proposes a low redundancy with acceptable accuracy classification method with data augmentation and modularity of neural networks using a Hierarchical Neural Network Tree (HNN-Tree). This method can accurately distinguish between visually similar items with different characteristics, aiming to improve task management and efficiency. Our approach integrates Augmentplus to do image augmentation to enhance the image and mean softmax likelihood to classify super cluster reducing the complexity. The models were trained and validated on the datasets Flower-102 and Food-101, achieving test error of 0.012 and 0.082, respectively. Compared to other artificial intelligence in image classification methods, our work demonstrates acceptable result test error but with a minimal number of parameters in image classification tasks.