Proteins Classification Using An Improve Darknet-53 Deep Learning Model

Abstract

Nowadays, the quantity of protein sequences saved inside the central protein database from laboratories around the sector is continuously increasing. The purpose is that experimental shape elucidation is exertions extensive and may be very time-consuming. Therefore, we want an automatic device that may classify the protein. The increased number of softmax classifiers and leakyrelu activation layer is used instead of the softmax classifier in the original darknet53 model which originally is less in number. Therefore, modifications helped in improving the structure and parameters of the Darknet model. These results revealed that this technique can also efficiently extract multi-layer features from protein images, regardless of batch size, and with greater accuracy. This model has the greater performance with an accuracy of 94.94 percent. The meaning of the experiment provides insight that helps biologists and scientists build the overall protein structure.