1. Introduction

With the growth of the global population, agriculture has become increasingly important in contemporary society. As one of the cornerstones of human civilization, agriculture plays a key role in ensuring global food supply and promoting socio-economic development [1]. However, traditional agriculture faces several challenges, such as resource scarcity, environmental pollution, and low production efficiency, which undoubtedly threaten its sustainability [2]. Against this backdrop, optimizing the agricultural production process using advanced technology to enhance efficiency and sustainability has become an urgent task.

This paper delves into the application of hyperspectral imaging and intelligent sensors in agriculture, demonstrating how these technologies are driving the precise and intelligent development of agriculture. The article discusses not only the application of hyperspectral imaging in soil, plant growth, and pest and disease monitoring but also explores the critical role of intelligent sensors in agricultural advancement. Through case analyses of the integrated application of these technologies, such as data fusion and decision support systems, it showcases their potential and prospects for improving agricultural production efficiency and precision.

The study aims to highlight that the application of hyperspectral imaging technology and intelligent sensor systems in agriculture can not only enhance the efficiency and precision of agricultural production but also promote the sustainable development of agriculture, which has profound significance for human society and the global environment.

2. Overview of Hyperspectral Imaging Technology

2.1. The Basic Principles of Hyperspectral Imaging

The principle of Hyperspectral Imaging (HSI) is that a hyperspectral imaging system acquires one-dimensional or two-dimensional information of a target scene through one of three basic sampling methods (point scanning, line scanning, and area scanning) [3]. It combines three types of scanning modes (Staring imaging, Whiskbroom, and Pushbroom) to collect spectral data for each pixel in the target scene image, forming a three-dimensional hyperspectral data cube with x, y spatial dimensions and λ spectral dimension. The hyperspectral data cube is shown in Figure 1 [4].

Figure 1. Hyperspectral data cube [4]

Hyperspectral imaging systems can be divided into three types based on the image acquisition mode: transmission, reflection, and interaction; and into two types based on whether they include a light source component: active and passive. Taking the hyperspectral camera commonly used in smart agriculture as an example, it is a type of reflective passive hyperspectral imaging system, generally employing line scanning and Pushbroom systems. The main components include front optical components (focusing components, collimating components, etc.), spectral separation units (grating, filter, Filter wheel), two-dimensional detectors (CCD, CMOS), and a computer [5, 6]. Referencing Figure 2 for illustration, the process begins with the illumination of the sample tissue by an external light source.

Figure 2. Pushbroom hyperspectral imaging system [7].

This illumination results in light being reflected off the tissue, which is subsequently directed onto a precision slit via the primary lens system. This slit plays a critical role, permitting only a select segment of light, corresponding to a specific linear portion of the tissue, to proceed. Following this, the light traverses through a set of optical components designed for collimation and spectral division, utilizing mechanisms such as prisms or diffraction gratings. These elements serve to segregate the incoming light into a sequence of finely demarcated spectral wavelengths. Post-separation, these wavelengths are converged onto an array of sensors. Consequently, the spectral data for each discrete pixel range is cast onto the sensor array's columns, effectively translating the examined tissue's spatial layout into an intricately composed two-dimensional representation, distinguished by its unique spatial and spectral dimensions.

2.2. The Application of Hyperspectral Imaging in Agriculture

Hyperspectral imaging technology, as a powerful tool, is widely used in agriculture for soil analysis, crop growth monitoring, and pest and disease detection. It captures more detailed information than traditional imaging technologies, significantly enhancing the precision and efficiency of agricultural management.

Within Brazil's Paraná, a revolutionary inquiry led by Amanda Silveira Reis and her colleagues delved into analyzing the organic matter within Oxisol soil [8]. This endeavor employed an avant-garde hyperspectral imaging sensor, paired with elaborate multivariate regression techniques, scrutinizing 384 soil samples across various stratified echelons. Techniques like Principal Component Analysis and Linear Discriminant Analysis were harnessed, marrying spectral data with soil organic matter indices, identified through standard lab methods, within a Partial Least Squares Regression model. This meticulous study's outcomes, boasting a 0.75 determination coefficient and a 0.87 correlation coefficient, along with a 2.1 Residual Prediction Deviation, underscore the efficacy of this model in mapping SOM content disparities and spectral anomalies [8].

This inquiry not only validates the predictive prowess of hyperspectral imagery in charting variations in soil organic matter but also amplifies its utility in the non-invasive evaluation of soil properties. It signals a stride forward in leveraging advanced agricultural techniques to safeguard soil organic matter across diverse landscapes.

The initiative led by Jianwei Qin, in collaboration with ARS and NASA's KSC, marks a significant leap in agricultural technology, especially in monitoring plant health. This system, adept at recording both reflective and fluorescent signals across a broad spectrum, was rigorously tested within NASA's plant growth enclosures on a variety of vegetables, notably Dragon’s Beard, to ascertain the impact of divergent watering regimes. Through machine learning, this system not only achieved an accuracy rate surpassing 90% in pinpointing hydration deficits but also blazed a trail in preemptive agricultural health diagnostics [9].

In a separate vein, Zhitao Xiao, Kai Yin, and their collective have forged a path in accurately identifying agricultural pests, marrying hyperspectral imaging with cutting-edge deep learning. Their methodology, spotlighting a novel spectral feature extraction coupled with an attention mechanism, significantly bolsters pest management strategies [10]. Employing 3D convolution branches at varied resolutions, this approach finely tunes the analysis of spectral-spatial dynamics, enhancing pest detection accuracy. Their model, when pitted against a dataset comprising nine pest species, outshined conventional methods, heralding a new era in agricultural pest identification [10].

2.3. Hyperspectral Data Processing and Analysis Methods

2.3.1. Data Preprocessing

Within the realm of hyperspectral data preparation, three core methods stand paramount: the transformation via Standard Normal Variate (SNV), the application of Multiplicative Scatter Correction (MSC), and the process of normalization [11]. These pivotal steps are foundational in enhancing the analytical outcomes' precision and dependability. By adjusting to a normalized mean and variance, SNV methodically diminishes the influence of external, non-biological discrepancies across the spectral readings of each analysis, thus improving sample-to-sample comparison. On the other hand, MSC employs a linear regression strategy to amend spectral variations, effectively mitigating distortions stemming from variances in particle size and other scattering influences, thereby aligning the spectral data more closely with a predefined reference. Normalization, in its essence, recalibrates the dataset to a defined scale, ensuring an equitable representation from all spectral bands within the analytical process and averting the predominance of any band due to larger numerical values. These preprocessing measures not only elevate the data's integrity but also set the stage for further analytical endeavors, encompassing feature extraction, classification, and the development of analytical models [11].

2.3.2. Spectral Feature Extraction

The process of isolating essential features plays a pivotal role in streamlining the analysis and categorization of data. This procedure typically employs techniques like Principal Component Analysis (PCA), Locally Linear Embedding (LLE), and the Successive Projections Algorithm (SPA), all aimed at distilling pertinent insights from complex datasets [11]. Within this context, PCA operates by distilling data into its most significant variance directions, effectively compressing the dataset while striving to conserve the bulk of its informational content. This is achieved by calculating the data's covariance matrix to ascertain its primary variance directions before reorienting the dataset along these axes, thereby simplifying its complexity. Conversely, LLE prioritizes the maintenance of local data point interrelations as it transitions the dataset into a dimensionally reduced space, proving adept at navigating datasets with inherent nonlinear characteristics. SPA's objective centers around the meticulous selection of spectral bands that most accurately capture the dataset's variability. Through a systematic band selection process, SPA endeavors to trim down the dataset's dimensions without compromising essential spectral information [11].

2.3.3. Data Classification and Model Building

In the realm of analyzing hyperspectral imagery, the art of selecting the most relevant features is crucial. Utilizing sophisticated analytical techniques like Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) plays a vital role in the accurate classification and development of predictive analytics models [11]. The process initiates with deploying algorithms such as Principal Component Analysis (PCA) or Minimum Redundancy Maximum Relevance (mRMR) to pinpoint the spectral bands loaded with pivotal information. SVM is distinguished for its ability to delineate an ideal separating hyperplane within a multifaceted feature space, utilizing kernel functions to form non-linear boundaries that enhance classification efficacy. Conversely, KNN capitalizes on the principle of proximity for classification, determining labels by the closeness to the most similar instances within the dataset [11]. During the model development phase, crucial parameters, like the kernel type for SVM and the value of K in KNN, undergo rigorous optimization based on the training data to refine performance. The success of this model is then evaluated through key performance indicators, including accuracy and precision, to verify its effectiveness and application readiness [11].

3. The Role of Intelligent Sensor Systems in Agriculture

3.1. Definition and Characteristics of Intelligent Sensor Systems

Intelligent sensor systems integrate sensors, data processing capabilities, and communication functions to monitor the environment or specific parameters and respond accordingly. These systems utilize Internet of Things (IoT) technology to achieve real-time data collection and transmission through wireless sensor networks (WSN), supporting remote monitoring and intelligent decision-making. Intelligent sensor systems are characterized by a high degree of automation, dynamic data collection, real-time monitoring, and analysis capabilities. They can also be integrated with cloud computing and artificial intelligence technologies to enhance the accuracy and efficiency of decision support [12].

3.2. The Advantages of Intelligent Sensor Systems in Multi-Scale Agricultures

In the realm of precision agriculture, the deployment of advanced sensor technologies facilitates uninterrupted observation and analysis throughout the crop cultivation cycle, encompassing both growth metrics and yield outcomes. Highlighting this evolution, Triantafyllou et al. have outlined the architecture of a pioneering monitoring framework predicated on the integration of Internet of Things (IoT) principles and the dynamics of wireless sensor networks [13]. This framework was applied in a detailed study focused on the cultivation of saffron in Kozani, Greece, showcasing its effectiveness in agricultural oversight. Additionally, the initiative known as mySense is dedicated to streamlining data collection through a hierarchically organized technological ecosystem, comprising sensor nodes, networking capabilities, cloud-based data management, and user-centric software solutions. This structure is designed to facilitate cost-effective, scalable, and cohesive technological integration, thereby promoting broader application of monitoring systems within agriculture [13].

In a parallel development, Liu et al. embarked on the creation of a novel IoT-based monitoring system for agriculture, leveraging open-source hardware to ensure scalability and cost-efficiency [13]. This system is distinguished by its intelligent IoT gateway, which incorporates motion sensing for the nuanced collection of data and the management of devices from afar. Beyond the confines of crop health surveillance, these intelligent sensor arrays extend their utility to the meticulous tracking of environmental parameters—temperature, humidity, luminosity, gaseous concentrations, and pH levels, alongside targeted temperature regulation. Such capabilities are indispensable for optimizing conditions within controlled agricultural settings, underscoring the critical role of technology in advancing agricultural productivity and environmental stewardship. Tanha and others explored the advantages of intelligent sensor systems in conserving resources in multiscale agriculture. By employing remote sensing technology, AI-driven remote sensors, automated irrigation systems, and Unmanned Aerial Vehicles (UAVs), precise monitoring and management of crop growth conditions were achieved, effectively conserving water resources, fertilizers, and chemical pesticides [14]. For instance, the use of subsurface drip irrigation technology minimized water loss due to evaporation and runoff, directly delivering water to the crop roots. Meanwhile, the irrigation system, through soil moisture sensors and raindrop sensors, remotely controls the opening and closing of drip irrigation, ensuring the accuracy of irrigation and efficient use of water resources [13].

4. Integrated Application of Hyperspectral Imaging and Intelligent Sensor Systems

The integrated application of hyperspectral imaging and intelligent sensor systems can significantly enhance the accuracy of identifying the growth environment and status of crops, such as early detection of pests, diseases, or water stress by monitoring the spectral characteristics of plant leaves. Additionally, this integrated approach can optimize agricultural management practices such as irrigation and fertilization. Through precise control, it can improve crop yield and quality, achieving the goals of precision agriculture. This integration offers a holistic view of crop health and environmental conditions, enabling farmers and agricultural professionals to make informed decisions based on comprehensive data analysis. This approach not only supports sustainable agricultural practices by minimizing resource use and environmental impact but also contributes to the advancement of smart farming technologies.

4.1. Multiscale Data Fusion

Systems dedicated to the environmental monitoring in agriculture increasingly adopt data integration approaches, leveraging an array of sensors—including those for assessing temperature, humidity, soil moisture—and hyperspectral imaging technologies to gather comprehensive data across various scales [15]. These setups often incorporate wireless communication protocols, visual displays, and microcontroller units to effectively visualize data, thereby informing agricultural practices such as watering and fertilization regimes. Insights from Prem's team highlight the prowess of IoT-enabled smart sensors in tracking crucial agricultural parameters, encompassing humidity, temperature, and soil's physical makeup. Furthermore, these sensors offer automated assessments of soil's nitrogen levels, aiding in precise fertilizer application. The deployment of IoT gadgets alongside drones equipped with hyperspectral imaging capabilities proves pivotal in the meticulous observation of plant health issues, including diseases and infestations. The integration of such innovations promises to elevate efficiency in agricultural outputs while curtailing economic setbacks, marking a leap forward in the realms of precision agriculture and the promotion of eco-friendly farming methodologies [15].

4.2. Construction of Decision Support System

The construction of decision support systems relies on the results of data fusion to provide data-based agricultural management recommendations and forecasts. For example, research tested an Intelligent Embedded Fuzzy Decision Support System (IEFDSS), which demonstrated an accuracy rate in field tests that was 96% higher than existing methods, showcasing its efficiency over traditional approaches. Additionally, devices based on image processing have been proven effective in addressing the technical issues of image resolution and processing speed found in traditional imaging systems, with a reported Lin’s concordance correlation coefficient of 0.99, demonstrating high accuracy [15]. This indicates that integrating advanced technologies into decision support systems can significantly improve the precision and reliability of agricultural management strategies, leading to optimized outcomes and enhanced productivity.

4.3. Case Study

In their investigation, Misra and colleagues shed light on the significant influences of the Internet of Things (IoT), big data, and AI in revolutionizing agri-food systems [16]. Their analysis spans across the implementation of sensor technology for greenhouse management, the use of UAVs for acquiring hyperspectral data, the development of automated agricultural equipment, and the application of spectral imaging coupled with sensor technology for assessing food quality. By harmonizing IoT, big data, and AI within the agricultural and food sectors, the research advocates for a transition towards more intelligent farming practices. This paradigm shift is characterized by the strategic use of interconnected sensors and automated systems, which not only facilitates precise monitoring and management of agricultural operations but also leverages AI to predict future agricultural trends and outcomes, thus aiding in the formulation of informed decisions and strategies in agriculture [16].

Hooi and others explored the potential and current applications of IoT and AI based on hyperspectral imaging and smart sensors in microalgae cultivation, monitoring, and optimization. Microalgae cultivation requires precise monitoring and control of cultivation parameters like biomass concentration, pH value, light intensity, temperature, and water level in tanks [17]. Traditional methods involve significant manual labor and are prone to inaccuracies due to environmental factors. The research emphasized implementing IoT and machine learning techniques with hyperspectral imaging and smart sensors to overcome these challenges. The system achieved real-time monitoring of relevant parameters, optimized the cultivation process while reducing manual labor, and resulted in higher biomass productivity [17].

A groundbreaking framework for soil and agricultural yield foresight was introduced by Gilson's team, leveraging past environmental data [18]. This method integrates the use of multispectral imaging and intelligent sensing devices, employing machine learning for the development of prognostic tools. The intent behind this innovation is to boost farm output and operational efficiency, with a keen eye on ecological preservation. The model's efficacy was validated through the analysis of meteorological data collected between 2001 and 2015, specifically targeting wheat cultivation regions. Results unveiled a calibration error (RMSEC) of merely 0.20 tons per hectare, a cross-validation error (RMSECV) of 0.54 tons per hectare, and a high Pearson correlation (R^2) of 0.9189. Exceptionally high predictability for soil organic matter and clay content was demonstrated, with R^2 values of 0.9345 and 0.9239, respectively, and an RMSECV of 0.54% for organic matter and 5.28% for clay [18]. This research highlights the profound implications of synthesizing sensor data and analytical models in agriculture, underscoring the transformative potential of IoT-driven precision farming in enhancing soil productivity assessments.

5. Conclusion

This study systematically analyzes the application of Hyperspectral Imaging Technology (HSI) and intelligent sensor systems in multiscale agriculture, highlighting the potential of these technologies in enhancing agricultural production efficiency and sustainability. By integrating HSI and smart sensors, precise monitoring of critical agricultural production areas such as soil, crop growth, and pest and disease incidence can be achieved, thereby optimizing resource use, reducing waste, and improving crop yield and quality. Moreover, the effective processing and analysis of hyperspectral data, along with the real-time monitoring and accuracy of smart sensor systems, provide agricultural management capabilities and decision support. While these technologies offer significant advantages in the agricultural sector, they still face challenges such as technical hurdles, complexity in data processing, and maintenance costs of hardware and software. Particularly, further research and innovation are needed in data fusion, construction of decision support systems, and integrated application in real-world scenarios. Future research directions should focus on addressing these challenges and exploring new integrated application models. Additionally, continuing to explore innovative applications of HSI and smart sensor systems in precision agriculture, resource management, and environmental monitoring will provide strong technological support for achieving a more efficient and sustainable agricultural production system.