Food processing refers to the techniques employed to convert raw materials into a product that can be fit for consumption. It is vital to ensure that the nutritional characteristics of raw materials are preserved and that toxicity is not introduced into the food during processing.
With rising expectations from consumers all over the world for food products of a reasonable standard or quality, there has arisen a necessity for the Industry to automate the processes using machines and systems and move away from the traditional practices of performing labour-intensive tasks. By modernizing their production processes, businesses are expected to benefit hugely, owing to various benefits arising from the process that includes improved accuracy when it comes to observing quality control, which is due to the availability of computers with a proper feedback mechanism.
Machine Vision Systems
Computer vision has been employed in various sectors, including but not limited to transport, surveillance, agriculture, etc. Computer Vision acquires different type of image data and analyzes it. In agriculture, MVS are employed in quality and safety inspection, grading, crop monitoring, etc. In these cases, the MVS collect various parameters like shape, weight, texture, etc. These parameters cannot be accurately measured by human eyes. In this way, with the employment of MVS, the possibilities for human error can be minimized to a great extent.
Machine Vision Systems aka MVS are proactively used in industrial as well as research environments. It can utilize one or more than one camera to capture, evaluate and recognize an object. MVSs can recognize stationary as well as moving objects.
MVS comprises of two main components, which are as under:
i. Image Acquisition
This component includes different types of cameras and the acquisition of image determines the embedded information of the image. Entire process of the MVS is dependent on the Image Acquisition. A MVS may capture pictures in real time by combining photographs, movies and other 2-D and 3-D image technologies. These images are capable of being transmitted to the processing unit by radio waves, cables and wireless sensors. There are many methods to obtain images for the sake of image processing, which are discussed in the table below:-
Apart from these, there are several other imaging methods which are under R&D that are being deployed in the market, one of them being the THz Wave Imaging system, which is capable of capturing images of meals moving on a conveyor belt.
1. Image Processing:
Image Processing is generally carried out by specialized image analyzing software. This plays a key role in the multitasking operations of MVSs. Image processing is used to create new pictures and image data from current photos. Image processing’s purpose is to isolate and enhance a region of interest in the body. This is a digital signal processing method and the computer does not attempt to understand the image’s meaning.
Image Processing has 3 stages:
(a) Low Level Processing
Low Level Processing is used to prepare the picture for analysis. Image Acquisition is a type of low-level processing. The captured image may not be a perfect match for the model that is being worked on and must be improved using methods such as contrast enhancement and noise reduction, much like any other ordinary photo editing programs.
(b) Intermediate Level Processing
Image operations at the intermediate level include image description, image representation and picture segmentation. Image segmentation is a necessary step in image processing. Image segmentation isolates the target from distracting items in an image. It improves the recognition system’s integrity and accuracy. Watershed Segmentation algorithm, K-Means Clustering and other techniques are used for image segmentation.
(c) High Level Processing
Image recognition and interpretation are examples of High-Level Image Processing. For this aim, many techniques such as K-Nearest Neighbour (KNN), Neural Networks, Support Vector Machine (SVM) and others are used.
Categorization of Foods
In Food Processing, some procedures are necessary to identify and categorize foods that are of high-quality, as well as sub-standard foods. Some of these procedures are as under:
(a) Food Safety and Quality Testing
Food Safety is vital to ensure that the health of the consumers is protected from any issues associated with consumption of food. By following proper food safety practices, consumers can be assured that the food produced contains the essential nutrients that are required for the human body, including ensuring the absence of any toxic components in the food. Food Safety isn’t only linked to the fields of sciences, but requires an interdisciplinary approach, so that proper hygiene can be maintained. Traditional methods to achieve food safety are labour-intensive, time consuming and impractical. It is important for the Industry to look for new methods for ensuring the safety of various types of food products.
In Quality Testing, the morphological characteristics of food, including but not limited to size, shape, colour, etc. are used to grade food. These morphological characteristics of a food product define its price or value in the market. MVS can observe these morphological characteristics and can categorize on the basis of that. The size and shape are quantized during the image processing. One example of this is a model proposed by Naik and Patel (2017), which used ripeness and size of a mango variety named “langda” as the key criteria for classification. Here, ripeness is calculated with the help of mean intensity algorithm performed upon L*a*b* colour space where (L* stands for Lightness and a*, b* are chromaticity coordinates).
There are various other algorithms that can be applied in Food Safety and Quality Testing like Gray Level Co-occurrence matrix and Gabor Filter.
(b) Monitoring and Packaging
Food Processing consists of numerous steps, rules and regulations that are required to be adhered to, in order to regulate the food quality as well as to ensure safety of workers. Taking the example of processing of a food which involves pressure change, this is required to be observed by staff from the Industry, which increases the probability of human error. Automation and MVS systems will help in optimizing the quality of food.
Foods are required to be packaged and transported from one place to another. Food packaging serves four purposes: sanitation, food safety, product display and transit convenience. In food packaging, human error is conceivable. Consequently, technology such as Machine Vision Systems may be utilized to monitor the process associated with food manufacturing and packaging.
(c) Foreign Object Detection
The presence of foreign objects in food could be one of the major reasons for a food product to be rejected by consumers. Food contamination caused due to the presence of any foreign objects might harm the reputation of the brand and could also have the potential of destroying brand loyalty among the consumers towards the brand’s products. Examples of foreign object contamination particularly for food products include gravels, insects, etc. With the use of machinery that is upgraded with advanced technology as well as with the rising levels of mechanization deployed in the Food Industry, the frequency of appearance of foreign objects in food-based products has been declining. Detection of foreign objects, which may be smaller in size is harder to detect with naked eye. MVS can become a helping hand and will assist in the separation of any foreign object from the food. Employment of these procedures ensures less wastage of resources and increases productivity.
Deep Learning Approaches
Machine learning functions by attempting to discover a pattern. It employs training samples to detect patterns and makes accurate predictions about future events. With increased data complexity and the development of Machine Learning techniques, classical Machine Learning becomes inadequate. Deep Learning Algorithms have evolved to accommodate increasingly complicated structures and greater data analysis capabilities.
Traditional Machine Learning included acting manually based on sample data sets and balancing the validity of the operation’s findings with the model’s interpretability. Based on the operating technique, traditional machine learning may be separated into supervised learning, unsupervised learning and reinforcement learning. Conventional Machine Learning methods rely on statistics as a reliable source. Following are several algorithms:
Deep Learning Methods
Deep Learning is a subset of machine learning that involves training Artificial Neural Networks with multiple layers to learn and extract increasingly abstract and complex features from data, enabling the network to make more accurate predictions or classifications. It is also known as Deep Neural Network. The learning effectiveness of a machine was observed to be significantly higher when Deep Learning method was employed, as compared to the traditional Machine Learning Method.
Deep Learning utilizes representational learning to study and comprehend data via the use of algorithms. In the case of MVS, observations or pictures can be represented in a variety of ways, such as a vector of intensity for each pixel, a sequence of edges or a form. Deep Learning frameworks of many sorts, including but not limited to Convolutional Neural Networks, Recurrent Neural Networks and Fully Convolutional Networks have been used in computer vision, speech recognition and other applications.
Deep Learning Frameworks
(a) Artificial Neural Networks (ANN)
An Artificial Neural Network is a computing system inspired by the structure of the human brain, consisting of interconnected nodes (neurons) that processes and transmits information. It is used to recognize patterns, make predictions or decisions and learn from data by adjusting the connections between neurons through a process called training.
(b) Convolutional Neural Networks (CNN)
A Convolutional Neural Network (CNN) is a type of artificial neural network commonly used in image and video analysis. It works by applying filters or “kernels” to small sections of input data, such as an image and using those filters to identify patterns and features in the data. The filters are then combined to produce a final output that represents the data in a way that is more meaningful for the task at hand, such as object recognition or image classification. The use of convolutional layers helps the network learn to recognize features in the input data, regardless of their location in the image, making it highly effective for image-based application.
(c) Fully Convolutional Networks (FCN)
A Fully Convolutional Network (FCN) is a type of neural network that is used for image segmentation, which involves dividing an image into segments and labelling each segment with a specific category. FCNs use convolutional layers to extract features from the input image and then use transposed convolutional layers to upsample the feature maps to the original image size. This allows the network to produce a pixel-wise classification of the image, where each pixel is labelled with a particular class. FCNs are commonly used in applications such as object detection, medical image analysis and autonomous driving.
Application of Deep Learning in Food Safety and Quality Evaluation
Hyperspectral Imaging Technique was applied to detect cold injury of peach and the Deep Learning framework utilized for this work was ANN to predict the quality parameters of the peach, (Pan et al., 2016). ANN was also adopted to classify the shapes of boiled shrimps and the overall accuracy achieved was 99.80%, (Poonnoy, et al., 2014). A MVS was designed to take tomato images and segment the Region of Interest. A Back Propagation Neural Network (BPNN) was used to classify maturity level of the tomato. (Wan et al., 2018)
Application of Deep Learning in Food Monitoring and Packaging
A CNN-based system was designed to check for all the details on food packaging labels. It used K-means clustering and KNN algorithm simultaneously, (Fabio et al., 2018). A neural network model was proposed to check for dates on food packages. Optical Character Recognition (OCR) and Optical Character Verification (OCV) were used in image segmentation and to extract the ROI and the result was fed to a FCN. (Ribeiro et al., 2018)
Application of Deep Learning in Foreign Object Detection
A model descriptwas proposed to detect and classify Fungi damages on commercial crops. The final prediction accuracy had turned out to be 86.48%, (Pujari et al., 2023). A model was designed based on CNN two detect metal parts in walnut fragments. The accuracy achieved from using this model was said to be 95%. (Rong et al., 2019)
i. Fabio, D. S. R., Francescom C., Mark, S., Kjartan, G., Georgios, L., Stefanos, K. (2018). An adaptable deep learning system for optical character verification in retail food packaging. Paper presented in 2018 IEEE International Conference on Evolving and Adaptive Intelligent Systems, EAIS, hodes during 25 – 27 May 2018.
ii. Naik, S., Patel, B. (2017). Thermal imaging with fuzzy classifier for maturity and size based non-destructive mango (Mangifera Indica L.) grading, Paper presented in 2017 International Conference on Emerging Trends and Innovation in ICT, Pune during 2 – 5 February 2017.
iii. Pan, L., Zhang, Q., Zhang, W/. Sun, Y., Hu, P., Tu, K. (2015). Detection of cold injury in peaches by hyperspectral reflectance imaging and artificial neural network, Food Chemistry, 192, 134-141.
iv. Poonnoy, P., Yodkeaw, P., Sriwai, A. Umongkol, P., Intamoon, S. (2014). Classification of Boiled Shrimp’s Shape Using Image Analysis and Artificial Neural Network Model, Journal of Food Process Engineering, 37(3), 257-263.
v. Pujari, J. D., Yakkundimath, R., Byadgu, A. S. (2023). Automatic fungal disease detection based on wavelet feature extraction and PCA analysis in commercial crops. Int. J. Image Graphics Signal Process, 6(1), 24-31.
vi. Ribeiro, F. D. S. et al. (2018). An End-to-End Deep Neural Architecture for Optical Character Verification and Recognition in Retail Food Packaging,” 2018 25th IEEE International Conference on Image Processing (ICIP), Athens, Greece, 2018, pp. 2376-2380, doi: 10.1109/ICIP.2018.8451555.
vii. Rong, D., Xie, L., Ying, Y. (2019). Computer vision detection of foreign objects in walnuts using deep learning, Computers and Electronics in Agriculture, 162, 1001-1010.
viii. Wan, P., Toudeshki, A., Tan, H., Ehsani, R. (2018). A methodology for fresh tomato maturity detection using computer vision, Computers and Electronics in Agriculture, 146, 43-50.
About the Authors:
The views/opinions expressed by authors on this website solely reflect the author(s) and do not necessarily reflect the views/opinions of the Editors/Publisher. Neither the Editors nor the Publisher can be held responsible and liable for consequences that may arise on account of errors/omissions appearing in the Articles/Opinions.