This paper addresses the issues of accuracy and efficiency in extracting the distribution areas of water hyacinth. A small-scale water hyacinth dataset was established based on drone-collected photos of water hyacinth in river channels, which was utilized for semantic segmentation tasks. A method based on deep learning was proposed to extract water hyacinth distribution areas from high-resolution drone remote sensing images. An efficient, accurate, and automated convolutional neural network called AttUNet was designed for this purpose, which eliminates the need for manually designed rules and can automatically learn remote sensing features of water hyacinth and extract distribution areas from images, thereby improving the accuracy and efficiency of acquiring relevant data. The research demonstrates that the proposed method can automatically extract features from massive high-resolution drone images, fully exploring complex nonlinear features, spectral features, and texture features in high-resolution drone images. The overall accuracy of extracting water hyacinth distribution areas in the study area reached 98.78%, with MIOU coefficient and mRecall coefficient of 95.86% and 98.01% respectively, both of which surpass the accuracy indicators of Deeplabv3+ and U-Net. The deep learning method for water hyacinth classification can fully exploit spectral, texture, and latent feature information in the data, and make it more suitable for extracting water hyacinth distribution information than traditional remote sensing classification methods.