Background and Objectives: Once the dynamic ECG signal based on single-lead is overwhelmed by noise during acquisition, the signal cannot be read. Recognizing and ignoring unreadable signal not only can reduce the misjudgment rate automatically analyzed by the analysis software but also can improve the efficiency of doctor's reading. In this article, clean signal and the signal which is disturbed but do not affect QRS discrimination are considered as readable signal, and a signal whose QRS is completely drowned by noise is regarded as unreadable. Materials and Methods：The data are obtained from partial electrocardiogram fragments of 37 real clinical patients, each fragment is divided into a segment of 4s, a total of 3,354 readable segments and 2,199 unreadable segments are obtained. Then process each segment as follows: 1) obtain sigNor by normalization; 2) opening-closing is performed to obtain openCloseSig; 3) Subtract the openCloseSig from sigNor to obtain the target signal. These operations suppress the large amplitude caused by motion interference and make the QRS more prominent. Then, Shannon entropy, kurtosis and other related characteristics were extracted. The random forest algorithm was used for classification and the model was validated by fivefold cross validation. Results: Compared with the model without morphological operator (92.94+/-0.93%), the accuracy of the model with morphology (85.68+/-1.30%) is improved obviously. In order to fully verify the effectiveness of the method, we use the "N" and "~" categories in the Physionet / Cinc Challenge 2017 data set for further verification, and the accuracy of adding morphological processing(93.75+/-0.69%) is also much higher than that of not adding (82.25+/-1.06%). Conclusion:In this paper, a method for recognizing unreadable segments of ECG signal based on morphological processing is proposed. Our results show that the addition of morphological processing can greatly help the recognition of unreadable signal.