Electroencephalography　(EEG)　is　a　non-invasive　method　for　measuring　brain　activity,　extensively　utilized　in　neuroscience　research.　The　process　of　reconstructing　potential　activation　sources　on　the　cortex　from　EEG　signals　measured　at　the　scalp　is　referred　to　as　EEG　Source　Imaging　(ESI).　Due　to　the　requirement　for　ESI　to　address　a　highly　ill-posed　inverse　problem,　traditional　methods　often　necessitate　the　design　of　neurophysiological　reasonable　priors　to　constrain　the　solution　space.　However,　it　is　difficult　to　design　a　neural　prior　that　accurately　reflects　the　properties　of　brain　sources.　To　overcome　this　limitation,　this　paper　proposes　a　hybrid　Long　Short-Term　Memory　(LSTM)　and　Transformer　network　for　ESI,　called　HLT-Net,　which　does　not　require　a　clear　definition　of　neural　priors.　More　specifically,　bidirectional　LSTM　is　introduced　to　capture　temporal　information　in　EEG　signals.　Then,　we　adopted　the　multi-head　attention　mechanism　in　the　Transformer　to　enhance　the　global　information　perception　of　model.　Furthermore,　a　mask　layer　has　been　added　to　the　input　of　the　model　to　enhance　its　robustness.　The　results　from　both　simulated　and　real　datasets　demonstrate　that　HLT-Net　outperforms　existing　technologies.　©　2024　IEEE.