Rapid and accurate recognition of crop leaf grade and disease is integral to the advancement of intelligent equipment for promoting refined management of agricultural products. In response to the problems of low accuracy and high cost of crop leaf grade and disease recognition, a crop leaf grade and disease recognition network (CLGDRNet) was proposed based on backbone information sharing and multi-receptive field feature adaptive fusion. Firstly, CSPNet, GhostNet and ShuffleNet were utilized to build a feature extraction backbone, and the feature information extracted by CSPNet, GhostNet and ShuffleNet was shared to achieve the purpose of information complementarity. Secondly, a multi-receptive field feature adaptive fusion module (MRFA) was designed, and the different receptive field feature maps were adaptively weighted and fused to highlight the effective channel information while enhancing the local receptive fields. Finally, an efficient multi-scale attention mechanism with deep gradient cross-space learning (EMAD) was proposed, the EMAD was embedded in the neck to obtain the deep gradient information and the target coordinate information, in addition, the context information of different scales was fused across the space, which could generate more accurate pixel-level attention to the deep feature map. The experimental results showed that the recognition accuracy of mAP@0.5 and mAP@0.5:0.95 for tobacco leaf grading dataset (TLGD) achieved 85.0% and 76.1%, respectively, and 97.6% and 74.2% for apple leaf disease dataset (ALDD), respectively. Compared with a variety of advanced target detection algorithms, CLGDRNet achieved higher recognition accuracy and faster recognition speed, which could provide key technical support for high-precision fine recognition of crop leaves.