How AI Analytics is Transforming Fruit Quality Control and Monitoring

• Food production systems’ enormous volumes of data must be precisely analyzed in real-time.
• AI analytics uses one or more AI tools. It extracts valuable information from big data on food quality, providing answers in an easy-to-understand manner during onsite food analysis.
• Three AI methods- machine/computer vision, machine learning, and deep learning are used for AI analytics in the food industry, for quality control from the farm to the retailers.

Fruit quality control and monitoring are necessary to ensure food safety, meet consumer demands, reduce labor and storage costs, and minimize food waste. Traditional methods are time-consuming and need instrumentation. Therefore, AI analytics in portable and stationary devices is quickly becoming standard for food evaluation. This article discusses the AI analytics used for fruit control and monitoring.

What is AI Analytics

Artificial Intelligence (AI) analytics uses one or many branches of AI to evaluate data and predict trends. It combines AI with traditional data analytics, where the analytical models or algorithms at the core of AI can analyze vast amounts of structured and unstructured data.

The huge data volumes are collected in various formats, kinds, and content. The data that has to be analyzed onsite can be heterogeneous (or multivariate) and have noise and redundancy. Organizing and cleaning data is necessary for traditional data analytics and requires time. Nowadays, AI tools that are automated can save time for these tasks.

Complex statistical models and computing systems, which form part of AI analytics, are capable of conditioning and preprocessing data and performing the analysis needed to extract relevant information from the data.

Before a model is used, it has to be trained or calibrated. One data set is used in training models, and a smaller one is used to test the models in the validation process. The results from the models are also compared to those from standard techniques to evaluate the model to judge if it can perform analysis independently onsite.

The variety of calibration data will influence a model’s prediction powers. For example, Near Infrared spectroscopy data on quality parameters like soluble sugars for an apple cultivar “Gala” from only one season and a few orchards will have less variation, so the predicting power of the model trained with it will be lower. A model trained with data from apples from many years and many sites with varying environmental conditions makes predictions more accurate and applicable for Gala apple testing in a commercial tool.

By making real-time analysis possible, AI Analytics is improving decision-making and enhancing operational efficiency in the entire supply chain of fruits.

Figure 1: Branches of Artificial Intelligence, Sharma et al. 2021. (Image credits: International Journal of Research in Engineering and Technology 8(8):2395-0056)

AI Analytics in the Agri-Food Sector

Agriculture operations are increasingly data-driven, from the farms and post-harvest stages to the retailers and food processing. None of the data is helpful unless it is analyzed in real-time and the insights’ presented are simple to understand so that they can guide users in making decisions.

AI is made of many branches (see Figure 1). The most well-known machine learning, computer vision, and deep learning are commonly employed in food production. AI tools can be used individually but are increasingly used in various combinations to tackle more complex tasks. Availability of cloud storage, smart machines, IoT, and blockchain technologies support AI analytics implementation in agri-food systems.

Figure 2: “AI Analytics applications for quality control of food,” Zhu et al. (2021). (Image credits:  https://doi.org/10.1016/j.crfs.2021.03.009)

There are numerous uses for AI analytics in food production. Precision agriculture could use a combination of machine learning, computer vision, and vegetative indices to analyze remotely sensed imagery to provide information on the health, growth, and productivity of crops and orchards. Field sensors that are part of IoT collect data on soil moisture and temperature, which can be combined with weather data from field installations or local weather stations to suggest irrigation schedules.

As fruit growers broaden their focus from only quantity to include quality and food safety, the uses of AI analytics increase. The following sections will discuss the different AI branches used in fruit quality control, starting from the farms.

Machine Learning

Machine learning requires manual pre-extracted data for analysis, like NIR spectral data.

Machine learning has three types: supervised, unsupervised, and reinforcement learning, depending on the training method and whether the data has labels.

Statistics, both qualitative and quantitive kinds are central to machine learning. The models can be pattern classification, probability estimation, regression analyses, etc.

The standard statistical models used in machine learning are Bayesian, Decision Tree, Support Vector Machine (SVM), Artificial Neural Networks (ANN), Fuzzy C Means, Random Forests, Convolutional neural network (CNN), K-Nearest Neighbors (KNN), Logistic Regression, and Linear Discriminate Analysis (LDA).

• Support Vector Machine is a Supervised learning model.
• K-means and Fuzzy C Means are unsupervised learning models.
• Random Forests and Decision Trees are pattern recognition models.
• Bayesian, Logistic Regression is a probability, estimation-based model.
• Principal Components Analysis (PCA) and Linear Discriminant Analysis (LDA) are dimensionality reduction methods.

The data analyzed can be qualitative or quantitative. Qualitative analysis can be helpful in detecting the shape, color, size, or odor of fruits and vegetables, while quantitive quality analysis will examine attributes like total soluble sugars, dry matter, titrable acidity, or pH.

Application

Machine learning models are often combined with Machine Vision Systems for various purposes:

1. Classification, evaluation of fruit grades, predicting trends, detection of bacterial spoilage, and quality.
2. It is also possible to use machine learning to analyze hyperspectral images from machine vision to estimate internal chemical quality parameters. The processing time is swift, and machine learning models can be integrated with real-time conveyors to grade and sort fruits. Or for use in portable devices for retailers and suppliers.
3. Machine learning models are also standard in monitoring and controlling food processing to enhance end-product quality.
4. Detecting food spoilage in packages for culling to maintain the quality of the batch.
5. Detecting foreign objects like fungal growth or insects through X-ray imaging and supervised learning models.

The foods monitored so far are apple, apricot, blueberry, cherry, citrus, mango, peach, pear, pomegranate, strawberry, tomato, onion, bean, coffee bean, olive, tomato juice, and walnut.

Deep learning

Figure 3: “The relationship between artificial intelligence, machine learning, and deep learning,” Zhu et al. (2021). (Image credits:https://doi.org/10.1016/j.crfs.2021.03.009)

Deep learning is part of machine learning, as Figure 3 shows. It is, however, more advanced than machine learning. Deep learning models involve deep neural networks and are suitable where multiple layers of functions are needed. These models are, however, not as easy as machine learning models to interpret.

Standard deep learning models are Artificial neural networks (ANN), Convolutional neural networks (CNN), Fully Convolutional Networks, and Recurrent Neural Networks. See Figure 4 for a diagrammatic depiction of CNN analysis.

Figure 4: “The architecture of a typical CNN,” Zhu et al. (2021). (Image credits:https://doi.org/10.1016/j.crfs.2021.03.009)

Application

Deep learning models have been used for food safety and quality evaluation applications when combined with machine vision. The latter detected the object, for example, honey, while ANN predicted the quality parameters of honey like antioxidant activity, total phenolic content, ash content, etc.

Deep learning models are used for quality control, food processing, monitoring, and detecting foreign objects. These models have been used for several vegetables, fruits, oils, retail food, and nuts.

Machine Vision Systems

Figure 5: “Components of a machine vision system,” Zhu et al. (2021). (Image credits:https://doi.org/10.1016/j.crfs.2021.03.009)

Machine vision systems allow for non-destructive and objective food evaluation and involve two steps: acquiring images and processing them. For this, a machine vision system needs cameras and image processing software used on a computer or machine, as shown in Figure 5.

One or more cameras acquire 2-D or 3-D images of moving or stationary objects. Images can be depth information through stereo systems, remote sensing photos, hyperspectral, X-ray, thermal, and magnetic resonance imaging.

Image processing can be low-level, intermediate-level, or high-level processing. Low-level processing preprocesses images. Intermediate-level processing covers image segmentation, representation, and description. High-level processing includes image interpretation and recognition and uses deep learning methods.

Applications include crop monitoring during harvest for fruit selection, quality, and safety inspection, produce grading and sorting, and foreign object detection. The parameters typically covered are food size, shape, color, texture, and weight. Machine vision systems exist for tomatoes, apples, mangoes, oranges, strawberries, biscuits, and irregular shape food.

Automated systems can capture details that humans can miss, and the work is repetitive, so it is ideal for machines.

Machine or computer vision systems grade fruits based on color, diameter, and discoloration caused by spoilage microbes. Fruits are then sorted to produce batches of similar maturity and size and for culling. Food selection for processing also uses machine vision systems.

Chemometrics for NIR Spectral Data

Chemometrics utilizing AI tools for NIR spectral data analysis for chemical quality parameters typically use machine learning methods, including ANN, since the data analyzed is not very large. Deep learning methods are more optimal only when large data sets need to be explored. Attempts made since 2016 to use deep learning models for chemometrics analysis of food have not demonstrated any improvement in analysis over classical methods.

Portable devices like quality meters manufactured by Felix Instruments Applied Food Science that can non-destructively measure total soluble solids, dry matter, color, and titrable acidity in real-time use ANN for analysis.

AI Analytics for Precision Quality Control

AI analytics allows the application of precision agriculture methods for quality control. By analyzing big data of various kinds and providing relevant information about fruits, AI ensures efficient harvesting, grading, sorting, packaging, processing, storage, and retailing. As a result, there will be less food loss, and fruit production will become more sustainable.

Sources

Admin. (n.d.). AI analytics. Virtusa. Retrieved from https://www.virtusa.com/digital-themes/ai-analytics

 

Elbir, Z., Caferoglu, B. A., & Cihan, O. (2022). Freshness grading of agricultural products using artificial intelligence. Artificial Intelligence Applications in Agriculture and Food Quality Improvement, 29–54. https://doi.org/10.4018/978-1-6684-5141-0.ch003

 

Maiyar, L. M., Ramanathan, R., Swamy, S. L., & Ramanathan, U. (2023). Fighting Food Waste: How Can Artificial Intelligence and analytics help? Innovation Analytics, 171–189. https://doi.org/10.1142/9781786349989_0008

 

Sahni, V., Srivastava, S., &  Khan, R. (2021). Modelling techniques to improve the quality of food using artificial intelligence. Journal of Food Quality, 2021, 1–10. https://doi.org/10.1155/2021/2140010

 

Sharma, S., Bisoyee, P., and Jathar, J. ( 2021). Quality Assurance using Artificial Intelligence: A Review Paper. International Journal of Research in Engineering and Technology 8(8):2395-0056

 

Sharma, S., Gahlawat, V. K., Rahul, K., Mor, R. S., & Malik, M. (2021). Sustainable innovations in the food industry through Artificial Intelligence and big data analytics. Logistics, 5(4), 66. https://doi.org/10.3390/logistics5040066

 

Sun, Z., & Stranieri, A. (2021). The nature of intelligent analytics. In Intelligent analytics with advanced multi-industry applications (pp. 1-21). IGI Global.

 

Zhu, L., Spachos, P., Pensini, E., & Plataniotis, K. N. (2021). Deep learning and machine vision for food processing: A survey. Current Research in Food Science, 4, 233-249. https://doi.org/10.1016/j.crfs.2021.03.009