内容摘要

A new artificial intelligence (AI) model can detect stress in agricultural crops using a nondestructive imaging technique, allowing changes in crop health to be monitored over time. Developed by a team of researchers at Penn State, the new model shows promise for accurately detecting crop stress caused by drought, disease, nutrient deficiency and other causes while using less computing power than some existing AI methods.

全文

Phys.org · 农业科研

Tested on mushrooms, an AI model shows potential for detecting crop stress with far fewer computations

Gaby Clark

Scientific Editor

Robert Egan

Senior Editor

A new artificial intelligence (AI) model can detect stress in agricultural crops using a nondestructive imaging technique, allowing changes in crop health to be monitored over time. Developed by a team of researchers at Penn State, the new model shows promise for accurately detecting crop stress caused by drought, disease, nutrient deficiency and other causes while using less computing power than some existing AI methods.

The researchers recently published their findings in Scientific Reports.

Three pathways reduce the computing burden

Processing large amounts of data from hyperspectral imaging—which involves using a special camera to capture information from many wavelengths of light—that measures changes in agricultural products and crops across time spans can be computationally expensive in terms of time, memory and energy, according to study senior author Shirin Ghatrehsamani, assistant professor of agricultural and biological engineering in the College of Agricultural Sciences.

"To address this problem, we developed 'PSUMamba,' which consists of three separate processing pathways," she said. "The first pathway analyzes spectral information to identify changes in crop biochemical properties, the second pathway examines changes in crop conditions over time, and the third pathway analyzes spatial features and patterns within crop images. These three pathways are combined to improve the model's ability to identify crop stress."

The researchers chose to evaluate the model in white button mushrooms because Pennsylvania grows 74% of the country's Agaricus—white button—mushrooms, and national mushroom sales fell 16% to $922 million last season, according to the U.S. Department of Agriculture (USDA). Mushrooms are economically important to Pennsylvania, and postharvest quality degradation represents a significant challenge for the industry, USDA reported.

One of the major challenges in modern agriculture is detecting stress early enough to prevent losses in yield and quality, said study first author Weiqun Wang, a doctoral candidate in the Department of Agricultural and Biological Engineering. Crops often show subtle physiological changes long before symptoms become visible to the human eye, he added, and hyperspectral imaging can capture this information.

"But the large volume of data generated makes analysis computationally intensive and difficult to deploy in practical agricultural settings," Wang said. "Our work addresses this challenge by developing an AI model that can accurately identify stress while requiring substantially fewer computational resources. This is important because it moves advanced sensing technologies closer to real-world use, where decisions often need to be made quickly and efficiently."

Testing stress in white button mushrooms

The team subjected white button mushrooms to controlled stress treatments to induce physiological deterioration representative of postharvest quality degradation. The experiment was conducted in the greenhouse facility of the Department of Agricultural and Biological Engineering.

Three types of stress were employed: thermal shock via aqueous immersion at 140 degrees Fahrenheit for 30 seconds, simulating heat exposure during processing; microwave radiation for 30 seconds, inducing localized cellular disruption; and mechanical injury through punctures, abrasion and impact damage, replicating physical handling stress. Untreated controls were maintained under identical environmental conditions without intervention.

High accuracy with fewer computations

The researchers used 204 bands of light collected from the mushrooms across eight timepoints. The model achieved an average accuracy of 94.47%, a sensitivity of 98.98% and an AUC-ROC—a standard performance metric used to evaluate classification models—of 99.36%.

The team reported that the model also demonstrated good computational efficiency, using approximately 10 times fewer computational operations to process data and produce predictions than the Vision Transformer model used for comparison. That deep learning model adapts AI transformer architecture, originally built for natural language processing, to process and analyze images.

The team also conducted experiments to evaluate the contribution of each pathway to the model's overall performance. Their results showed that spectral, temporal and spatial information provide complementary features and that combining all three pathways improves how well the model distinguishes among different categories of crop stress indicators, based on the information it receives.

Wang noted that in the AI world, Mamba is a type of deep learning neural network designed to process information efficiently, particularly long sequences of data. The main contribution of this study, he pointed out, is the development of a new AI architecture that merges three important types of information from hyperspectral images.

Field trials remain the next test

"PSUMamba can detect agricultural stress with high accuracy while maintaining relatively low computational requirements," he said. "More research is needed to determine if the model will work under real agricultural field conditions with different crops and products."

Although this study was conducted using controlled stress treatments, the broader implication extends beyond a single crop or product, Ghatrehsamani pointed out. The framework was designed to capture fundamental stress-related patterns, which opens the possibility of adapting it to many agricultural applications, including monitoring disease, drought stress, nutrient deficiencies and postharvest quality changes.

Early and reliable stress detection could help producers make more informed management decisions, reduce input costs, minimize waste and improve sustainability, she explained.

"The next step is to evaluate the model under real agricultural conditions and across a wider range of crops and stress factors," Ghatrehsamani said. "We are interested in testing whether the approach can be integrated with field-based sensing systems, greenhouse-monitoring platforms, robotic systems and potentially aerial-imaging technologies. Ultimately, our goal is to develop intelligent sensing tools that can provide growers with timely, actionable information to support precision agriculture and improve crop productivity, quality and resilience."

Publication details

Weiqun Wang et al., Triple-path bidirectional Mamba enables linear-complexity temporal hyperspectral plant stress classification, Scientific Reports (2026). DOI: 10.1038/s41598-026-74809-5

Journal information:
Scientific Reports

Provided by
Pennsylvania State University

Citation: Tested on mushrooms, an AI model shows potential for detecting crop stress with far fewer computations (2026, November 20) retrieved 20 November 2026 from https://phys.org/news/2026-11-mushrooms-ai-potential-crop-stress.html

了解完整信息

Phys.org · 农业科研
前往原文

继续探索