Abstract: Pesticide residue detection methods are prone to interference from spatial noise in leaf images, resulting in low detection accuracy．Therefore, a method for detecting pesticide residues on surface of crop leaves based on discrete wavelet transform features was proposed．Using a multimodal fusion network to extract surface feature vectors of crop leaves, an attention module was introduced in network model to perform asymmetric modal fusion processing on feature vectors to eliminate spatial noise．By filtering and correcting edges of leaf image, leaf image edges were segmented, and original image was decomposed and transformed step by step using a discrete wavelet transform feature algorithm, and spectral reflectance of leaf surface was calculated．Based on this, area of interest on surface of leaf was determined, and characteristic spectrum processing method was used to distinguish spectra of pesticides, in order to achieve pesticide residue detection．Comparative experimental results showed that, the proposed method has high detection accuracy for pesticide residue detection on surface of crop leaves．