Scientists in Israel have created an AI-based machine-learning model using hyperspectral imaging to assess tomato quality before harvest. The “non-destructive” method predicts key quality traits like weight, firmness and lycopene content, helping farmers monitor fruit development in real time, optimize harvest timing and enhance crop quality.

Hyperspectral images of specific ranges of light wavelengths, known as “spectral bands,” are used to study objects’ properties based on how they reflect light. Calling it a “leap forward in precision agriculture,” the team believes this approach addresses challenges associated with traditional methods, offering a faster and cost-effective alternative.

“Traditional methods for assessing tomato quality are time-consuming, destructive and expensive, limiting their use to small post-harvest samples. Farmers lack practical tools for pre-harvest quality monitoring, which is essential for optimizing growing conditions and improving crop quality,” lead author Dr. David Helman, head of the modeling and monitoring vegetation systems lab at the Hebrew University of Jerusalem, which conducted the study, tells Food Ingredients First.

“Existing non-destructive technologies, like hyperspectral cameras, are costly and impractical for field use.”

Hyperspectral imaging can help farmers monitor tomato quality “throughout the ripening process, enabling them to adjust management practices and determine the optimal harvest time, which ensures better fruit quality.”

Besides improving the nutritional quality, the team’s AI-driven technology could also allow for “better adaptation to environmental changes, enhance the resilience of agricultural systems and contribute to global food security,” he adds.

Tapping machine learning

Hebrew University conducted the study, published in Computers and Electronics in Agriculture, in collaboration with researchers from Bar-Ilan University and the Volcani Center.

The teams used a handheld hyperspectral camera to gather data from 567 tomatoes across five varieties. Machine learning algorithms, like Random Forest and Artificial Neural Networks, predicted seven key traits: weight, firmness, TSS, citric acid, ascorbic acid, lycopene and pH.

“This data is processed through [machine learning] algorithms, such as Random Forest and Artificial Neural Networks, to predict key quality parameters like weight, firmness, lycopene content and acidity,” explains Helman. Total soluble solids and pH of tomatoes were the other traits analyzed.

“The model identifies correlations between the spectral data and these parameters, enabling non-destructive quality assessments in real-time.”

The team observed “high accuracy” in these models. Random Forest achieved an R² of 0.94 for weight and 0.89 for firmness, which means it predicted tomato weight and firmness with 94% and 89% accuracy respectively. 

Eyeing efficiency and scale

Helman notes that the research aims to bridge the gap between advanced imaging technology, AI and practical agricultural applications.

“This work has the potential to revolutionize quality monitoring not only in tomatoes but also in other crops. Our next step is to build a low-cost device (ToMAI-SENS) based on our model that will be used across the fruit value chain, from farms to consumers.”

“The ToMAI-SENS model is planned to be used with a compact, affordable and field-friendly device compared to traditional hyperspectral cameras or NIR spectrometers, which are expensive and impractical for widespread agricultural use.”

The model also analyzes multiple fruits simultaneously for “efficient and scalable” high-throughput assessments, he tells us.

The team plans to conduct rigorous validation tests and explore partnerships for commercialization for ToMAI-SENS.

Impacting agriculture policies

AI-driven tools like ToMAI-SENS could shape policies by providing data-driven insights that enhance crop monitoring and quality assurance, underscores Helman.

“This could lead to more targeted subsidies, encouraging sustainable farming practices and efficient resource use. Governments might prioritize funding for technologies that improve food security and quality.”

The team is currently seeking industry partners to support further R&D “with opportunities for intellectual property licensing,” he shares.

Lab-to-market

Heman reveals that the researchers aim to expand the method’s application and integrate it into automated harvesting systems.

“Collaborating with manufacturers will enable the integration of this technology into existing agricultural tools and potentially expand its application to other crops.”

In future, smart harvesting systems and consumer tools could integrate the machine learning model to evaluate produce quality in supermarkets, the study notes.

Early feedback for the team’s fruit quality monitoring tool has indicated interest in its potential for “cost-effective and scalable quality monitoring.”

“Farmers and food companies see value in their ability to track quality pre-harvest and streamline processes. The tool’s affordability and usability are expected to resonate strongly with agricultural stakeholders as well as day-to-day consumers,” he concludes.