Automatic inspection of flow pack packaging using AI

Flow pack packaging

Until now, automatic inspection of quality defects in flow pack packaging was practically impossible, due to their wide variety of shapes, film colours and the different products contained. 

This type of packaging is very common in the food industry, and has a multitude of possible finishes and sizes or shapes, which made automatic inspection practically impossible using conventional artificial vision techniques to detect the main quality defects (welding, product and labeling).

Thanks to INSPECTRA’s Easy Ai ® artificial intelligence, it is now possible to detect most quality defects in this type of packaging regardless of the type of packaged product or the finish of the film.

What is AI inspection?

Deep learning, is a branch of artificial intelligence that uses artificial neural networks to learn to perform complex image classification tasks.

When we carry out packaging inspection, a large part of the welding defects are usually detectable through image analysis, which can be captured in different lighting conditions.

When these defects have a tone that clearly contrasts with the welding area, they are usually detected by comparing the color tones, grouping the areas of a certain tone into regions, performing the KO classification if said region has a specific size or shape. determined.

On other occasions, defects in the welding area do not have a characteristic color, and in this case color tone comparison techniques usually generate many false positives, which leads many producers to carry out the inspection manually assuming a high operating cost in production, making it impossible to automate the end of the line.

When the welding defect is distinguishable for a human being by analyzing the image, its detection becomes possible using artificial intelligence by training a convolutional neural network.

For example, a lack of adhesion defect has a similar tone to that of a correct weld, for this reason it cannot be detected by comparison of shades. However, a neural network is capable of learning to identify it, showing an intense red hue in the defect on a heat map. A neural network heat map is a visual representation of the importance of each pixel in an image in making a classification decision, and serves to validate whether the neural network’s classification criteria are accurate or not.

A neural network has the advantage that it can learn to detect new defects or to do so in new types of packaging, being able to teach it continuously and unlimitedly as the packaging changes.

Inspectra's Easy AI ® technology

At INSPECTRA we have Easy AI ® artificial intelligence technology, which allows us to detect any quality defect that is perceptible to humans in images captured by our equipment, capable of capturing images of processes that happen at high speed and in spectrums beyond the visible, facilitating the entire process of launching inspection applications through artificial intelligence.

The process that goes from the creation to the launch of a new inspection program using artificial intelligence is carried out in three different phases: capture and labeling, training and start-up.

INSPECTRA offers optional image processing using artificial intelligence across its entire range of equipment. To facilitate the creation of inspection programs with artificial intelligence, it has a set of tools that provide clients with autonomy when creating new inspection programs, which we discuss below:



INSPECTRA equipment is prepared to record the set of images necessary for training a neural network using the following tools:

    • (ISS) Equipment interface: ISS in its interface allows the option of saving 100% of images in different directories to facilitate subsequent classification in order to obtain the training dataset
    • (IPB) Management of production batches: IPB allows the copy of images from the equipment to the training PC, facilitating their extraction for each production batch, and can be easily saved in folders.
    • Labeling Software: We also provide our clients with different work tools for labeling images, so that batches of images can be easily reviewed and labeled in a very short time.


INSPECTRA provides, together with its equipment, an artificial intelligence tool for training and validating the learning of neural networks. Our DEEP LEARNING software allows anyone with basic computer skills to train neural networks.

The software has the following functionalities:

    • Intuitive, easy-to-use user interface
    • Image bank augmentation through image pre-processing
    • Selection of pre-trained neural networks of different complexity and number of layers
    • Quick workouts
    • Unlimited learning.
    • Visualization of results with heat maps to validate neural networks
    • Automatic generation of results reports


INSPECTRA equipment allows the rapid start-up of a neural network, allowing the start-up of new inspection programs in a few hours, providing the following advantages:

    • Loading new trained networks in two steps (copy and click) on all INSPECTRA devices
    • Processed in REAL TIME LOCALLY without the need for a cloud connection
    • Full control over your own data, it is not shared with external servers
    • No inference/training costs or annual license
    • Supervision of operation in production using IPB software

Below, we show some typical examples of quality failures detectable through artificial intelligence in packaging with defects that are impossible to detect for conventional artificial vision equipment.

Quality defects in packaging detectable by artificial intelligence

Complex defects in welding

    • Product trapping. Sometimes part of the packaged product, or pieces of it, are trapped in the welding area and this causes poor sealing.
  • Incorrect sealing temperature in the packaging machine, sometimes the temperature exceeds what is necessary or is insufficient, which causes the welding to be defective. This defect usually manifests itself at the junction of the longitudinal weld with the transverse weld of the container.
  • Film displacement: Sometimes the film moves from its correct position and the silkscreen on the container is not positioned correctly.


These types of defects are usually detectable by code readers, but as soon as the background of the product is not homogeneous or the print is deformed due to the curvature of the film, artificial intelligence is required to read it. The LABEL CHECKER software of the INSPECTRA equipment incorporates the functionality of label reading with artificial intelligence, allowing the extraction of texts and codes in the most difficult conditions. The most common errors in printing codes are the following:

    • Programming error. When a change in production occurs, it is necessary to update the printing programs of the labeling machines so that they print the codes that correspond to the product currently in production.
    • Print head failure: On other occasions, the container is not flat, or the print head runs out of ink or is damaged, resulting in poor printing or the complete absence of printing.


On other occasions, the packaged product has a defect that is noticeable through the packaging.

    • Broken product: When the packaged product has broken or deteriorated and is noticeable through some part of the packaging film.
    • Contamination: When it is possible to detect foreign bodies or some type of contamination through some part of the packaging film
    • Poor disposition: When the product has not been disposed correctly in the packaging.
    • Incorrect quantity of product: When there are more or less units in the container (usually it is detected on a scale, but in certain cases it is not detectable due to weight difference)

AI detection examples

Below we show different images of success stories of quality inspection using artificial intelligence of packaged products.


Displacement of the film in packaging

code printing defects


Do you want to know more?