基于自适应特征融合改进AlexNet的水稻磷素营养诊断

打开文本图片集

中图分类号：TP391.41；S511 文献标志码：A

Abstract： 【Objective】 To diagnose the phosphorus （P） nutritional status of rice （Oryza sativa L.） more accurately and eficiently，and improve rice yield and quality. 【Method】 A phosphorus nutrition diagnosis method for rice was proposed based on an improved AlexNet model enhanced with adaptive feature fusion. The AlexNet was used as the backbone network. First， a transfer learning strategy was adopted by initializing the base network with pre-trained weights obtained from the ImageNet dataset， thereby enhancing initial feature representationcapability.Additionally，aresidual moduleandan Inception module wereembedded intoeachof the four feature extraction stages to improve deep local feature learning and multi-scale feature representation. Finally， an adaptive feature fusion mechanism was introduced to dynamically weight and effectively integrate features from different layers，enhancing the model sensitivity to key phosphorus-related information. 【Result】 The improved AlexNet model achieved the classification accuracies of 94.81% and 86.35% at the tillerig and jointing stages of rice respectively，representing improvements of 7.44 and 20.77 percentage points compared to the original AlexNet. Compared with AlexNet， GhostNet， and ResNet34，the improved AlexNet model demonstrated superior performance with the precision of 94.86% and the recall of 94.81% at the tillering stage，while at the jointing stage， the precision and recall were 86.30% and 86.35% respectively. On the public Plant Village dataset for plant disease clasification，the improved AlexNet model also achieved excellent performance， with the accuracy of 99.24% ，the precision of 99.25% ， and the recall of 99.24% ，which demonstrated its effectivenessand generalization capability.【Conclusion】This improved AlexNet model enables precise and eficient diagnosisof phosphorus deficiency in rice，thereby providing atheoretical support for scientific fertilization of rice.Furthermore，the model serves as arobust scientific reference for disease status diagnosis in other crops.

Key words： Rice; Phosphorus nutrition; AlexNet; Residual module; Inception module; Adaptive feature fusion; Transfer learning

在水稻OryzasativaL.的生长发育过程中，氮、磷、钾等营养元素扮演着至关重要的角色。（剩余17635字）