Against the backdrop of the rapid development of smart agriculture, pest and disease monitoring and crop growth assessment for large-scale farmlands are of substantial importance for precision management and risk early warning. However, traditional unimodal visual methods are highly susceptible to illumination variation, canopy occlusion, scale differences, and background interference in real field environments, and thus fail to make full use of environmental sensing information and spatial priors. To address these issues, a multimodal target perception framework for intelligent farmland inspection is proposed in this study. By jointly integrating UAV imagery, time-series data from ground Internet of Things sensors, and spatial positional information, joint modeling of pest and disease recognition and crop growth assessment is achieved through cross-modal alignment and collaborative encoding, multi-scale target perception, and dynamic multimodal fusion and decision-making. Experimental results demonstrate that, in the pest and disease recognition task, the proposed method achieved a Precision of $91.63\%$, a Recall of $90.27\%$, an F1-score of $90.94\%$, and an mAP of $93.15\%$, significantly outperforming comparison models such as Faster R-CNN with ResNet50 backbone, YOLOv8-m, Swin Transformer-Tiny, and Multimodal Transformer. In the crop growth assessment task, an Accuracy of $89.96\%$, a Precision of $89.11\%$, a Recall of $88.74\%$, and a Macro-F1 of $88.92\%$ were achieved, again clearly exceeding those of ResNet50, EfficientNet-B3, ViT-B/16, and conventional multimodal fusion models. The ablation study further verified the effectiveness of the cross-modal alignment module, the multi-scale target perception module, and the dynamic fusion module, with the complete model reaching $90.94\%$, $93.15\%$, and $88.92\%$ in Pest F1, Pest mAP, and Growth Macro-F1, respectively. Furthermore, the net economic return regression experiment at the unit-area level further demonstrates that the proposed method can effectively connect state information with economic outcomes, showing strong application potential in return prediction, performance evaluation, and resource allocation optimization. These findings indicate that the proposed method can effectively improve perception accuracy and robustness in complex farmland environments, thereby providing reliable technical support for intelligent inspection, pest and disease early warning, and precision management in agricultural scenarios.

This study was conducted to address the demand for interpretable intelligent recognition of fruit tree diseases in smart horticultural environments. A KAD-Former framework integrating an agricultural knowledge graph with a visual Transformer was proposed and systematically validated through extensive cross-regional, multi-variety, and multi-disease experiments. The primary objective of this work was to overcome the limitations of conventional deep models, including insufficient interpretability, unstable recognition of weak disease features, and poor cross-regional generalization. In the experimental evaluation, the model achieved significant advantages across multiple representative tasks: in the overall performance comparison, KAD-Former reached an accuracy of $0.946$, an F1-score of $0.933$, and a mAP of $0.938$, outperforming classical models such as ResNet50, EfficientNet, and Swin-T. In the cross-regional generalization assessment, a DGS of $0.933$ was obtained, notably surpassing competing models. In terms of explainability consistency, a Consistency@5 score of $0.826$ indicated strong alignment between the model’s attention regions and expert annotations. The ablation experiments further demonstrated that the three core modules—AKG (agricultural knowledge graph), SAM (semantic alignment module), and KGA (knowledge-guided attention)—each contributed substantially to final performance, with the complete model exhibiting the best results. These findings collectively demonstrate the comprehensive advantages of KAD-Former in disease classification, symptom localization, model interpretability, and cross-domain transfer. The proposed method not only achieved state-of-the-art performance in pure visual tasks but also advanced knowledge-enhanced and interpretable reasoning by emulating the diagnostic logic employed by agricultural experts in real orchard scenarios. Through the integration of the agricultural knowledge graph, semantic alignment, and knowledge-guided attention, the model maintained stable performance under challenging conditions such as complex illumination, background noise, and weak lesion features, while exhibiting strong robustness in cross-region and cross-variety transfer tests. Furthermore, the experimental results indicated that the approach enhanced fine-grained recognition capabilities for various fruit tree diseases, including apple ring rot, brown spot, powdery mildew, and downy mildew.