Dutta Pijush, Dey Amit, Das Raushan, Maji Raghunath, Paul Shobhondeb, Mandal Sudip
Greater Kolkata College of Engineering and Management, Baruipur, West Bengal, India.
Data Scientist,TCS Oval, Newtown, West Bengal, India.
Methods Mol Biol. 2025;2952:297-313. doi: 10.1007/978-1-0716-4690-8_18.
Traditional machine learning methods have been replaced by deep learning (DL), a modern, cutting-edge approach to classifying textures and localizing tissues. This study introduces a dataset for texture classification at the image level using integrated transfer learning-based EfficientNet-B7 deep convolutional neural networks (CNN). The dataset comprised 381 images of dimensions 150 × 150 pixels for training, validation, and testing. The model attained an accuracy of 89.33%, 52.43%, and, 51.326% for training, validation, and testing datasets while training losses are 0.2513, 1.846, and 1.6137, respectively. The straightforwardness of the framework, setup, and execution also plays a role in the findings of this research. Moreover, these proposed techniques can suggest extracting more prognostic information than an experienced human observer.
传统的机器学习方法已被深度学习（DL）所取代，深度学习是一种用于纹理分类和组织定位的现代前沿方法。本研究引入了一个数据集，用于在图像级别使用基于集成迁移学习的高效神经网络（EfficientNet-B7）深度卷积神经网络（CNN）进行纹理分类。该数据集包含381张尺寸为150×150像素的图像，用于训练、验证和测试。该模型在训练、验证和测试数据集上的准确率分别达到89.33%、52.43%和51.326%，而训练损失分别为0.2513、1.846和1.6137。该框架、设置和执行的简易性也对本研究的结果起到了作用。此外，这些提出的技术能够比经验丰富的人类观察者提取更多的预后信息。
