This paper presents the research results given by a convolutional neural network trained with continuous Wavelet transform (CWT) images for the classification of six hand grip postures with EMG signals obtained through an 8-channel MYO ARMBAND. These data were obtained by performing the respective extraction of EMG signals to ten patients who have a hand amputation. The procedure used for the extraction and preprocessing of adequate data is demonstrated, with which accuracy achieved a yield of up to 94.73% for the correct selection of the hand grip postures.