This paper introduces a quantum convolutional neural network (QCNN) inspired by classical convolutional neural networks (CNNs) for quantum machine learning. The QCNN uses only O(log N) variational parameters for N qubits, enabling efficient training and implementation on near-term quantum devices. It combines multi-scale entanglement renormalization ansatz (MERA) and quantum error correction. The QCNN is applied to two tasks: recognizing quantum states of 1D symmetry-protected topological (SPT) phases and optimizing quantum error correction (QEC) codes. For SPT phase recognition, the QCNN accurately identifies phases by learning to distinguish between different quantum states. For QEC, the QCNN optimizes both encoding and decoding procedures, outperforming known codes. The QCNN's structure is similar to CNNs, with convolution and pooling layers followed by a fully connected layer. It is shown that the QCNN can recognize SPT phases with high accuracy and reduce sample complexity for phase detection. The QCNN is also used to design optimized QEC codes, which perform better than existing codes in certain error models. The paper discusses potential experimental implementations and generalizations of QCNNs. The QCNN's architecture is efficient, scalable, and suitable for near-term quantum devices. It provides a promising approach for quantum machine learning tasks, including phase recognition and error correction.This paper introduces a quantum convolutional neural network (QCNN) inspired by classical convolutional neural networks (CNNs) for quantum machine learning. The QCNN uses only O(log N) variational parameters for N qubits, enabling efficient training and implementation on near-term quantum devices. It combines multi-scale entanglement renormalization ansatz (MERA) and quantum error correction. The QCNN is applied to two tasks: recognizing quantum states of 1D symmetry-protected topological (SPT) phases and optimizing quantum error correction (QEC) codes. For SPT phase recognition, the QCNN accurately identifies phases by learning to distinguish between different quantum states. For QEC, the QCNN optimizes both encoding and decoding procedures, outperforming known codes. The QCNN's structure is similar to CNNs, with convolution and pooling layers followed by a fully connected layer. It is shown that the QCNN can recognize SPT phases with high accuracy and reduce sample complexity for phase detection. The QCNN is also used to design optimized QEC codes, which perform better than existing codes in certain error models. The paper discusses potential experimental implementations and generalizations of QCNNs. The QCNN's architecture is efficient, scalable, and suitable for near-term quantum devices. It provides a promising approach for quantum machine learning tasks, including phase recognition and error correction.