Ureteropelvic　Junction　Obstruction　(UPJO)　is　a　common　hydronephrosis　disease　in　children　that　can　result　in　even　progressive　loss　of　renal　function.　Ultrasonography　as　a　preliminary　diagnostic　step　for　UPJO　has　the　nature　of　economical,　radiationless,　noninvasive,　and　high-noise.　Artificial　intelligence　has　been　widely　applied　to　medical　fields　and　can　greatly　assistant　for　doctors＇　diagnostic　ability.　We　build　and　test　a　DWT-utilized　classifier　for　UPJO　diagnosis　using　ultrasound　images.　Our　diagnosis　model　is　a　combination　of　an　attention-based　pyramid　semantic　segmentation　network　and　a　discrete　wavelet　transformation　processed　residual　classification　network.　We　also　compare　the　performance　between　benchmark　models　and　our　models.　Our　diagnosis　model　outperformed　benchmarks　on　classification　task　with　accuracy=91.77%.　This　model　can　automatically　grade　the　severity　of　UPJO　by　ultrasound　images,　assistant　for　doctors＇　diagnostic　ability,　and　relieve　patients＇　burden.　©　2022　IEEE.