Abstract: 【Objective】This study aimed to systematically improve the YOLOv5n model from two aspects （network structure and loss function） to comprehensively enhance the degree of model lightweighting and detection accuracy for complex rice leaf diseases， providing technical support for the intelligent recognition of rice leaf diseases， thus meeting the demands of intelligent management for modern precision agriculture.【Method】Based on the YOLOv5n model， the CBRM module， MLCA module， and ShuffleBlock module were first introduced into the backbone network to reconstruct the original network structure. Subsequently， the Shuffle Attention mechanism was embedded before each C3 layer of the neck network to facilitate collaborative feature learning in both channel and spatial dimensions. Finally， the original CIoU loss function was replaced with the EIoU loss function to improve disease spot localization accuracy and enhance the convergence stability of the model during training. To validate the effectiveness of the proposed method， experimental evaluations were conducted on an image dataset comprising three major types of rice leaf diseases： bacterial leaf spot， brown spot， and leaf blast.【Result】The YOLOv5n-SHCA model demonstrated stable recognition of characteristics of the three rice leaf diseases （bacterial leaf spet，brown spot， and leaf blast） and exhibited high sensitivity in detecting small-target disease spots. In comparative tests with the mainstream YOLO model， the YOLOv5n-SHCA model achieved effective improvements in multiple performance indicators， while substantially reducing the parameter scale， reflecting a favorable balance between detection accuracy and lightweight design， showing strong potential for practical application. Specifically， the model attained a mAP@50 of 0.966， a mAP@50-95 of 0.746， a precision of 94.2%， a recall of 92.0%， an inference speed of 1.4 ms/image， and a parameter size reduced to 1.53 M. The detection accuracies for bacterial leaf spot， brown spot， and leaf blast were 0.994， 0.921， and 0.983 respectively， all outperforming the YOLOv5n model. The proposed model effectively addressed the problems of complex factors such as dense disease spots， small disease spot scales， and morphological similarity of rice leaf blast on brown spot detection， thereby enhancing perception and discrimination of disease spot features of the model under complex background noise.【Conclusion】By constructing a lightweight backbone network comprising the CBRM module， MLCA module， and ShuffleBlock module， integrating the Shuffle Attention mechanism into the neck network， and adopting the EIoU loss function in place of CIoU， the proposed YOLOv5n-SHCA model effectively improves the detection accuracy of complex rice leaf diseases while achieving model lightweigh-ting， which provides technical support for the intelligent recognition of rice leaf diseases.