Aiming at the problems of low accuracy and small number of samples in field amaranth identification, the YOLO v5 amaranth identification model was improved by introducing ASPP attention mechanism of expanding receptive field and extracting context information. The improved model would significantly improve F1 value and mAP index under low data set. The experimental results showed that the F1 value and mAP of amaranth identification model was increased by 13 percentage points and 18.6 percentage points after the introduction of ASPP attention mechanism in low data set. The detection rate of amaranth was increased by 15.4 percentage points with horizontal recording. Therefore, the research provided an effective method for the identification of amaranth or other weeds under low data sets, and prepared for the research of weed identification and management in the agricultural field.