Triatomines are vectors of Trypanosoma cruzi, the parasite that causes Chagas disease. Bug identification is a key factor to control T. cruzi transmission because triatomines have different vectorial capacities. There is a great diversity of insects that frequently invade houses and can be misidentified as triatomines. New tools have been developed to facilitate triatomine identification, such as electronic keys (TriatoKey, TriatoDex) and automatic identification systems that use artificial intelligence. So far, they have been developed to identify adult bugs only. Identification of triatomine nymphs is also necessary because whenever nymphs are present, there is evidence of colonization and greater risk of T. cruzi transmission. As already demonstrated for adult bugs, a promising artificial intelligence technique to identify triatomine nymphs is deep learning. Convolutional Neural Network is a deep learning-based algorithm that learns to differentiate classes of objects through training with images previously classified. In this work we implemented an algorithm that uses deep learning with a pretrained network called AlexNet. We investigate the accuracy of AlexNet to identify: 1) Nymphs vs adults of hemipterans; 2) Nymphs of triatomines vs non-triatomines; 3) Triatomine nymphs vs bed bugs; and 4) Triatoma, Rhodnius and Panstrongylus nymphs. We provided to the algorithm 2587 images from websites and photographs of specimens available at University of Brasília and Fiocruz-Bahia. The images had different color backgrounds, textures, and positions (dorsal and dorsolateral). From the overall number of images, 80% was used for training the AlexNet algorithm while 10% was used for testing and the rest 10% for validation. After training, we obtained confusion matrices/heatmaps with the amount of classification errors and the accuracy (probability of correct identifications) for each group. We calculated the accuracy with a 95% confidence interval around the means of 10 simulations. In the first analysis (nymph vs adult hemipterans), we observed a mean accuracy of 92.3% (CI 95% 86-96). AlexNet accuracy was higher to identify triatomine vs non-triatomine nymphs (98.5%, CI 95% 88-99). We observed a better result to identify triatomine nymphs vs bed bugs, 100% accuracy (CI 95% 94-100). However, the identification of triatomine genera showed an average accuracy of 92.4% (CI 95% 83-97); we observed that the confidence interval was wide for Panstrongylus (CI 95% 43-90) and Rhodnius (CI 95% 60-96) nymphs. AlexNet demonstrated high performance to identify nymphs and adults of hemipterans, nymphs of triatomines and non-triatomines (including bed bugs). However, nymph identification of different genera of triatomines (Triatoma, Rhodnius, and Panstrongylus) showed lower accuracy values and the performance of AlexNet may be improved by increasing the quality and the quantity of images. We conclude that automatic identification of triatomine nymphs using a deep learning algorithm (AlexNet) is a valuable tool and strengthens similar identification systems developed for adult bugs for the surveillance of T. cruzi vectors.