Abstract: To improve speed and accuracy of fruit detection, a smart intelligent fruit detection method for agricultural robots based on improved YOLOv7 model was proposed．Backbone network of improved YOLOv7 model was RepVGG instead of CSPDarknet, and ECA（efficient channel attention）mechanism was introduced．Taking apple as research object, and adopting improved YOLOv7 model to fruit intelligent detection．Simulation results showed that the method could accurately achieve agricultural robot apple fruit detection, with detection accuracy, recall, precision, and average accuracy reaching 97.67%, 95.38%, 95.11%, and 93.17%, respectively, and had a fast detection speed．Detection time for each image was 11.21 ms．Compared with FPN, SSD, and Faster R-CNN model, improved YOLOv7 model had faster detection speed and higher detection accuracy, and could be used in practical applications of agricultural intelligent cloud monitoring, providing a reference for improving speed and accuracy of fruit detection．