<p>Edge detection is one of the core tasks in computer vision. Existing edge detection networks often fail to fully consider the feedback effect of deep information on shallow information during feature extraction and fusion. To overcome this limitation, we propose DFFNet (Deep Feature Feedback Network), an innovative edge detection model based on deep feature feedback. Specifically, we design three novel modules in the encoding network: the Deep Feature Fusion Module (DFFM), which utilizes dynamic convolution to effectively fuse multiple feature layers into a single feature layer; the Deep Feature Attention Module (DFAM), which combines channel and positional attention mechanisms to highlight key information; and the Deep Feature Enhancement Module (DFEM), which extracts semantic information through fully connected layers and feeds the enhanced deep features back to shallow layers to improve edge representation. Additionally, we adopt a dual decoding structure in the decoding network, integrating short and long connectivity designs, and propose a cross-scale fusion module that merges features of different scales through nearest-neighbor interpolation and feature concatenation. Experimental results show that DFFNet achieves a single-scale ODS F measure of 0.831 on the BSDS500 dataset and shows excellent performance on the Multicue public dataset, highlighting its strong potential for edge detection tasks. Our research provides new design insights for deep feature-driven edge detection models with broad application prospects.</p>