Artificial Intelligence techniques enabled insights into Leather Defects

With the advent of the digital revolution, the detection of leather surface defects has gained immense significance towards automation in the assessment of leather quality, which is of paramount importance in the leather trade that has eventually become global. The proposed work has strived to develop an artificial intelligence-enabled reliable and efficient system for detecting leather surface defects using a leather image dataset. The work has utilized conventional machine learning algorithms and deep learning approaches for distinguishing leather surfaces. However, it has been found that due to the variability in the leather surface and defects, the conventional machine learning algorithms have not been able to satisfactorily distinguish the leather surfaces. As a result, LeatherNet, a novel lightweight deep neural network, has been proposed. For better analysis, the performance of LeatherNet has been compared with the performances of prominent existing convolutional neural network models, previously experimented machine learning algorithms, and existing state-ofthe-art methods in this domain. The performance of LeatherNet has been found to outperform all the algorithms, architectures, and existing state-of-the-art methods considered. Accuracy, loss, precision, recall, and AUC score metrics have been used for performance measurement. When trained for 1500 epochs, the proposed model has recorded maximum training accuracy, precision, and recall of 99.78%, 99.69%, and 99.92% respectively, while the maximum testing accuracy, precision, and recall have been recorded at 97.42%, 97.66%, and 99.40% respectively.