Artificial intelligence-based approaches for Pb2+ detection in agricultural soils: A review

Lead ion (Pb2+) contamination in agricultural soils poses serious risks to ecosystem stability and human health due to its persistence and bioaccumulative toxicity. Traditional detection techniques are constrained by complex sample preparation, matrix interference, long analysis cycles, and limited in situ applicability. To overcome these limitations, artificial intelligence (AI)-based approaches have been increasingly adopted. Given the heterogeneity of agricultural soils and sensing signals, model selection is inherently scenario-specific: Support vector machine/support vector regression are suitable for complex matrices and multi-ion interference; artificial neural networks are preferred for low-concentration or multi-ion detection; convolutional neural networks are effective for high-dimensional, weak, or multi-modal signals; and least absolute shrinkage and selection operator/Ridge methods enable rapid, low-cost field screening. The performance of AI models depends on key factors—such as training dataset, hyperparameter optimization, and validation metrics. Coordinated optimization of these parameters enables robust, precise, and interpretable Pb2+ quantification. AI applications in soil Pb2+ detection are categorized into three main approaches: (i) single-modality approaches enhance sensitivity and specificity to address Pb2+ alloying and weak signal issues; (ii) multi-modal fusion strategies effectively mitigate interferences from complex soil matrices; and (iii) automated integrated platforms enable fast, field-deployable analyses while minimizing manual intervention. The approaches form a coherent technical chain that progressively addresses key bottlenecks in agricultural soil Pb2+ detection. Nevertheless, research in this area still faces challenges, including field adaptability, chemical speciation decoupling, and interference under variable soil conditions. Future research should focus on synergistic strategies that integrate materials, AI, and field applications. This review provides a targeted reference for the accurate tracking and control of Pb2+ contamination in farmland soils.