Welcome to Francis Academic Press

Academic Journal of Engineering and Technology Science, 2020, 3(6); doi: 10.25236/AJETS.2020.030610.

Research on the Robust and Low-cost Computer Method Based on Deep Residual Learning


Chung Ka Po

Corresponding Author:
Chung Ka Po

The Webb Schools, Claremont, California


At this exact moment on our planet, there's are more than 65.2 million people suffering from visual illnesses caused by cataract. It is the cause of a third of all blindness in the world, and 99% of these patients live in developing countries. However the recovery surgery is one of the most cost-effective ones in the field. Lack of medical care is the real cause of such skyrocketing number of patients in developing country. Out of this reason, I developed an integrated diagnosis system based on deep learning methods, which could diagnose cataract with the accuracy of 91.7%. The high accuracy and the low-cost features of this diagnosis method make it an excellent auxiliary tool of preliminary diagnosis in developing countries. The best performance is held by the 50-layer deep residual neural network trained with Adam optimizer, which could adjust learning rate according to the training status and specific weights. The further experiments with the quantity of data as a variable indicated that the best performance of deeper model is limited by the insufficient data.


Low-cost, Computer, Deep Residual Learning

Cite This Paper

Chung Ka Po. Research on the Robust and Low-cost Computer Method Based on Deep Residual Learning. Academic Journal of Engineering and Technology Science (2020) Vol. 3 Issue 6: 85-93. https://doi.org/10.25236/AJETS.2020.030610.


[1] Steinkuller, P. G. (1983). Cataract: The leading cause of blindness and vision loss in Africa. Social Science & Medicine, 17 (22), 1693-1702.
[2] Kingma, D. P., & Ba, J. (2014). Adam: A Method for Stochastic Optimization. arXiv: Learning,.
[3] Roelofs, G. H. (2005). History of the Portable Network Graphics (PNG) Format. Linux Journal.
[4] Ioffe, S., & Szegedy, C. (2015). Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. arXiv: Learning,.
[5] De Boer, P., Kroese, D. P., Mannor, S., & Rubinstein, R. Y. (2005). A Tutorial on the Cross-Entropy Method. Annals of Operations Research, 134 (1), 19-67.