Acoustic imaging is a key technology for applications such as noise source localization and abnormal sound diagnosis. Since the acoustic signals are non-modulated broadband signals, existing acoustic imaging methods divide microphone array data into several subbands and then perform acoustic imaging on each sub-band separately. However, the energy distribution of the acoustic signals across different frequency bands is uneven, leading to potential estimation errors in some sub-bands due to low signal-to-noise ratios, significantly impacting the accuracy of acoustic imaging. To address this issue, research was conducted on band-weighting methods based on complex Gaussian mixture models. By jointly utilizing data from multiple frequency bands to assign weights to each sub-band, the impact of subbands with erroneous estimates on the accuracy of acoustic imaging is reduced. To validate the effectiveness of the proposed method, experimental verification was conducted, measuring the accuracy of acoustic imaging using indicators such as the false alert rate, miss detection rate, and root mean square error. Experimental results demonstrate that the method effectively improves the accuracy of acoustic imaging, particularly reducing the false alert rate by more than 2. 1% under conditions where the signal-to-noise ratio below 10 dB.