Panoramic sensing of harmonic levels in distribution network is an important means to achieve power quality control. However, the distribution network has many points and is constrained by the high cost and operation and maintenance workload of power quality monitoring(PQM) devices, which makes it difficult to widely configure monitoring devices at each node to realize the panoramic sensing of harmonic levels. In order to solve the above problems, a harmonic level estimation model based on improved spatial and channel attention augmentation and densely connected network (SECBAM-Densenet) is proposed in this paper. Active power, reactive power, harmonic voltages, harmonic currents, and other electrical characteristics are collected by portable PQM at the grid-connected points that are not equipped with stationary PQM.The SECBAM-Densenet model is trained with the measured data to establish a nonlinear mapping relationship between the input characteristics and harmonic levels, and the correction matrix and constraint terms are constructed to improve the estimation effect of the model. The harmonic levels of the distribution network are effectively estimated in real time by applying the measured active power and reactive power of the electricity meter and other operating data.The example results verify the accuracy of the proposed estimation method and provide a feasible solution for harmonic sensing in distribution networks without PQM nodes.