Mental workload recognition is of great significance to improve the work performance of human-computer interaction operators. At present, some studies have shown that good classification results have been achieved by extracting the energy characteristics of electroencephalogram (EEG) signals for mental workload recognition. However, this method only focuses on the amplitude information of the signal, while ignoring the phase information. Only the frequency domain characteristics of each channel are analyzed, and the synchronization relationship between the signals of different channels is not considered. In order to fully consider the functional connectivity of different brain regions, a new method of mental workload classification based on weighted phase lag index (WPLI) heat map was proposed. The WPLI between the two channels of the preprocessed EEG signal was calculated and the heat map was drawn to evaluate the phase coupling between the signals of different channels, thus reflecting the functional connectivity of different brain regions. It can be intuitively observed from the WPLI heat map that there are significant differences in the distribution of brain functional connectivity under high and low load conditions. Through the experiments, the energy feature map and WPLI heat map were classified by histogram of oriented gradient-support vector machine (HOG-SVM) and LeNet-5 respectively. The results show that the combination of WPLI heat map and LeNet-5 has better classification results.