The　low-voltage　DC　power　system　is　widely　applied　in　solar　systems,　electric　vehicles,　large　data　centers,　etc.　DC　arc　fault　becomes　threaten　to　the　safe　operation　of　the　DC　systems.　When　the　arc　fault　is　in　different　states,　the　control　and　isolation　of　the　fault　lines　or　devices　should　be　different.　In　view　of　the　above　problems,　this　paper　proposes　an　approach　to　evaluate　the　DC　arc　fault　risk　degree　under　different　test　conditions　including　electrode　material,　air　pressure,　current,　etc.　based　on　back　propagation　(BP)　neural　network　algorithm.　Firstly,　the　Otsu　method　is　used　to　automatically　select　the　threshold　value　to　binarize　the　arc　images.　The　connected　region　is　traversed　to　denoise　the　binarized　images　to　improve　the　clarity　of　the　images.　Secondly,　the　images　are　transformed　into　Hue,　Saturation　and　Value　(HSV)　space　to　extract　the　color　features　of　the　images,　and　the　color　features　are　sent　to　the　classifier　constructed　by　the　BP　neural　network　to　train　the　features　to　realize　the　recognition　of　different　electrode　materials.　The　morphological　features　and　texture　features　of　the　images　are　extracted.　According　to　the　effective　features,　the　risk　degree　of　the　arc　fault　is　divided　into　three　levels.　The　effective　features　are　sent　into　the　classifier　to　train　the　features　to　evaluate　the　arc　fault　risk　degree　levels.　The　results　show　that　the　accuracy　of　the　algorithm　is　more　than　95%　for　the　evaluation　of　arc　fault　risk　degrees　and　electrode　material　identification.　This　method　has　a　significance　to　evaluate　arc　fault　risk.　©　2021　IEEE.