Diabetic retinopathy (DR) is a retinal pathology caused by diabetes, and is one of the main causes of blindness worldwide; Its early detection is essential in order to prevent its progression in the patient. There are various methods for early diagnosis, among these, it has been shown that convolutional neural networks (CNN) are suitable for the analysis of this phenomenon, contributing to the early diagnosis of this disease. In addition, deep learning techniques (DL -Deep Learning) have been used, the models proposed in the literature focus on the stages of preprocessing, extraction and selection of image features; however, these models may suffer from overfitting and the use of regularization techniques to control it has not been considered. So, in the present work, it has been proposed from a conceptual and experimental point of view, the selection of five regularization techniques on five pretrained deep learning models and through the analysis of metrics (precision, Recall, F1 score) a Artificial neural network regularization technique that improves the generalization capacity for the classification of diabetic retinopathy images. it is an abbreviated presentation. A maximum length of 250 words should be used. It is recommended that this summary be analytical, that is, that it be complete, with quantitative and qualitative information, generally including the following aspects: objectives, design, place and circumstances, patients (or objective of the study), intervention, measurements and main results, and conclusions. At the end of the summary, keywords taken from the text should be used, which allow the retrieval of information.