Food Industry Inspection: Foreign Objects and Quality

This Category positioning

In-line machine vision inspection of food product appearance, size and shape, color and surface, production process and packaging

In food industry visual inspection, the first hurdle is not the algorithm, but site conditions. The workshop needs washing, sterilization, and dust removal, so the equipment must be easy to clean and maintain, and its material, structure, and ingress protection rating must all be set according to the site's hygiene management regulations; this is entirely different from the requirements for placing a piece of vision equipment in an ordinary machining workshop. The equipment structure, lens and light source protection, and cable routing all need to be considered together during the solution design stage.

The second barrier is Inspected Object Form. Food items are irregular in shape, naturally vary, are often stacked or occluded, and their surfaces may be wet, oily, or highly reflective. Solutions therefore usually need singulation (separating stacked items), single-file alignment (stabilizing the inspection pose), and two-sided imaging, with illumination designed specifically for the surface characteristics. Products such as biscuits and potato chips also have Thin, brittle, light, irregularly shaped, which requires solving high-speed conveying, random orientation and overlapping at the same time.

The third barrier is Judgement and Traceability. For food, "conforming" is often tied to hygiene and safety: the critical size of foreign matter, the allowable range of appearance defects, and the distinction between normal baking texture and genuine defects all need to be quantitatively defined together with the quality control department; how NG products are handled (rejection, diversion, re-inspection, sample retention) must also be aligned with the on-site process, and inspection images and results need to be kept on file to support quality analysis.

inspection Pain Points

Why Food Production Lines Should Move from Manual Visual Inspection to Inline Visual Inspection

As food production lines move toward continuous, high-speed and automated operation, the traditional quality screening approach based on manual visual inspection faces challenges in inspection efficiency, judgement consistency and adaptability to production cycle time. These are concentrated in the following aspects:

Core Technical Solution

Broken down by "industry × product × inspection issue" into five combinable inspection directions

Food visual inspection is not suited to a simple split such as "one solution for candy, one for biscuits"; it is better to divide it by The process step where the inspection object sits divided into five directions. Each direction corresponds to a relatively fixed set of inspection scope and imaging approach, and an actual project is usually a combination of two or three of them.

inspection Scope

What can be inspected in each of the five directions

inspection Object

Ten product directions divided by food category

Product forms in the food industry vary widely, and the same "appearance inspection" has different imaging difficulties and inspection priorities across categories. The table below organizes ten product directions that can be planned independently by category, each with a different inspection focus:

The conveying and imaging structures differ markedly between directions. Single-file conveyed products (candy, biscuits, chocolate) suit fixed-pitch piece-by-piece inspection; bulk products (nuts, granules, chunks) usually use Vibratory bowl or conveyor spreading + high-speed area-scan / line-scan + AI classification + air-blow sorting approach, so that pieces are identified one by one, classified one by one and rejected one by one; for packaged products, images need to be taken at a station after packaging, taking into account the light transmission properties of the packaging material.

Typical defect

Which defects and anomalies these types of solutions mainly address

inspection Capabilities

What conventional vision and AI are each good at, and where the boundary lies

Not every food defect requires AI. For inspection items with clear rules, such as contour, dimension, color range, presence/absence, and count, conventional vision algorithms are usually easier to parameterize and easier to validate and explain; defects with complex forms, large variation in normal texture, or that are hard to describe with fixed rules are better handled with AI vision algorithms. In actual projects, the vast majority are Combination of Both.

AI vision algorithm

The algorithm chain from image acquisition to OK / NG judgement

Algorithms for food appearance inspection are usually organized as a fixed chain, with each step solving one class of problem and the software finally outputting a combined judgement:

• 01 Product positioning: determining the inspection region of each product in the image, solving the problems of position offset, spacing variation and multiple products entering the field of view at the same time
• 02 Target segmentation: separate products from the background, and assess the separability of touching or overlapping products
• 03 Contour and geometric analysis: extracts the outer contour and calculates geometric parameters such as dimensions, area, roundness and contour deviation
• 04 Surface and color analysis: gray level, color, texture, and zone analysis are used to identify anomalies with distinct image features such as stains, discoloration, and scorch marks
• 05 AI defect recognition: build recognition models for complex defect shapes, completing defect extraction and classification
• 06 Result fusion and judgement: merge the conclusions of each inspection item and output an overall OK / NG with the defect type

Typical AI Algorithm Tasks in Food Inspection

Vision Software Platform

The control layer that handles image processing, parameter management, result output and data statistics

Visual inspection software is not just an interface that displays images; it is one of the control layers of the whole system: it handles image processing and judgement logic, and also takes on recipe and parameter management, external communication of results, retention of defect images, and production data statistics. Food production lines usually involve switching between multiple product specifications, so the software needs to support fast switching of inspection recipes.

Technology Architecture

From product imaging to quality judgement, then to production line execution and data management

The complete system is organized in five layers, each with clear responsibilities and each able to be replaced or extended separately; on top of this, a unified AI algorithm platform and visual inspection software form a complete closed loop from "seeing a defect" to "automatic judgement, automatic rejection, and data traceability".

inspection Method

How to choose between 2D, 2.5D, 3D, line-scan and area-scan

The choice of imaging method depends on three factors: the type of feature that must be detected on the product, the production cycle time (conveying speed), and the range of height variation of the product. The following are the applicable scope and trade-off logic of common methods:

equipment and Complete Machine

Structure and site adaptation of food inspection machine models

The complete machine of food visual inspection equipment usually consists of a conveying section, a visual inspection section, a rejection / diversion section, and an electrical control section. The frame uses a structure that is easy to clean, and the electrical control and vision components are arranged separately from the conveying area for ease of daily cleaning and maintenance; the infeed and outfeed sections can be fitted with side guards and guides for product spacing and pose alignment.

Applicable Industry

Which Food Sub-Segments These Solutions Typically Serve

In terms of product form, the common applications of this type of solution include: baked goods (cookies, biscuits, wafers, bread, cake, egg tarts), candy and chocolate (hard candy, soft candy, filled candy, lollipops, chocolate blocks), nuts and roasted snacks (peanuts, sunflower seeds, almonds, cashews, pistachios, mixed nuts), puffed and snack foods (potato chips, shrimp crackers, rice crisps, corn flakes), fruits, vegetables and fresh-cut produce, and granular and block foods (jelly, preserved fruit, soy products, regular block meat products), as well as packaging forms such as bags, boxes, bottles, cans and blister packs.

Typical Case Studies

Reference for food inspection items organized according to the standard case structure

The two cases below are organized with a unified project structure: project scenario -> inspection requirement -> overall workflow -> technical solution -> algorithm and software -> rejection interlocking -> implementation priorities -> acceptance method. The cases illustrate how a solution is organized and how implementation is approached, The inspection items listed are those that may be involved within the scope of this project assessment; they do not mean that any site has confirmed the presence of these defects.

Case 01 | Visual Inspection of Both Sides on a Candy Production Line

Project Scenario: After forming, cooling and settling, the candy enters the visual inspection area on a conveyor belt. The system runs the candy through the complete sequence of "front-side inspection -> back-side inspection -> image matching -> AI defect recognition -> OK / NG -> automatic rejection", and conforming products move on to the downstream packaging process step.

Two technical approaches to double-sided inspection:

Result matching and judgement: After top and bottom vision, or two stations, output their results separately, the software associates them by product ID, trigger sequence, encoder position or time relationship. For example, if product A is "front OK + back OK", it is judged as a final OK; if product B is "front OK + back NG", it is judged as a final NG. If the candy is prone to sticking, deformation or has a coated surface, we recommend first validating a non-contact bottom-surface imaging method, with the final selection subject to on-site sample trial results.

Implementation Focus: ① Minimum defect size confirmation — for example, when a customer requires detection of a 0.5 mm defect, the field of view and resolution must be calculated to determine whether that defect can achieve sufficient pixel coverage; the specific resolution is calculated from the on-site product dimensions, field of view and speed, and verified with a sample trial; ② Optical stability — shield the inspection area from light and use a stable illumination structure to prevent changes in ambient light from affecting color and surface judgement; ③ Product position adaptability — use mechanical guides and spacing to improve the consistency with which products enter the vision area; ④ Front and back sides synchronous design — combine conveying speed, product spacing and trigger signals to ensure that the results on both sides correspond to the same product; ⑤ Defect sample validation — collect samples that are normal, chipped corner, crack, foreign matter, color abnormality, damage, abnormal shape, etc., and establish the final judgement rules based on the actual quality standard.

Case 02|Cookie Production Line Appearance Defect and Baking State Inspection

Project Scenario: For products such as biscuits, cookies, wafers and sandwich biscuits, machine vision performs inline inspection before baking, cooling, arranging and packaging. Unlike candy, the criteria for biscuits also include Color, shape, cracks, missing corners, baking state and surface texture, so the solution focuses on distinguishing "normal baking texture" from "real defects".

Typical solution chain: Conveyor → pitch spacing → upper vision → lower vision (optional) → AI algorithm → sorting. If only top-side inspection is required, High-speed area-scan vision completes the main inspection items; if the bottom surface must also be inspected, top-and-bottom dual vision or flip-over inspection is used.

Algorithm Task Allocation: Traditional vision is suited to outer contour, dimension, color, missing corner, and missing part inspection; AI vision is suited to cracks, baking abnormalities, complex surface defects, and irregular damage, forming a chain of "positioning → segmentation → defect extraction → defect classification → OK / NG". Implementation Challenges: Cookies themselves have natural baking texture, embossing, holes or regular patterns, and these normal features may produce image appearances similar to cracks and scorch marks, so during algorithm development samples must be collected of normal products, of different batches, colors and baking states, of each type of real defect, and of boundary cases between normal texture and suspected defects, in order to reduce false calls; in addition, cookies are fragile food, so the rejection mechanism must be evaluated for air-blow pressure, pusher speed and contact method to avoid secondary breakage.

Implementation Workflow

The complete path from requirement discussion to on-line acceptance

• 01 Project requirement discussion: confirm the product type, production line structure, cycle time and inspection purpose
• 02 Product sample analysis: analyze the shape, material, color, surface characteristics and conveying state
• 03 Vision feasibility assessment: judge whether the imaging method and algorithm route are feasible, and identify risk items
• 04 Defect standard review: define defect types, acceptance criteria and how to handle edge cases together with quality control
• 05 Vision sample trial: run illumination experiments and algorithm validation on real samples, and output measured conclusions
• 06 Equipment solution design: determine the camera, lens, light source, mechanism, and control method
• 07 System integration and joint debugging: interfacing with PLCs, rejection mechanisms and data interfaces
• 08 On-site validation and acceptance: test coverage, stability and rejection accuracy against the confirmed samples

The focus of project implementation is never just choosing a camera; it is solving at the same time Product imaging, defect recognition, motion synchronization and automatic rejection Four stages. Of these, sample trial validation is the step most easily skipped, yet it affects downstream acceptance the most — a visible defect does not mean the defect can be classified reliably; inspection capability must be confirmed through sample testing and on-site validation.

Why Choose EEK

Turnkey delivery covering optical imaging, vision algorithms, AI defect recognition and automation integration

EEK eeK focuses on machine vision inspection solutions, centering on Optical imaging, vision algorithms, AI defect recognition, visual inspection software and automation equipment integration, providing integrated solutions from vision validation to in-line mass production inspection for industries such as food, automotive parts, stamped parts, and electronic hardware.

For the food industry, EEK combines 2D vision, 2.5D vision, 3D vision, and AI vision algorithms to perform inline inspection of food appearance defects, color, dimension, shape, presence/absence, packaging, and the production process, and connects PLCs, rejection mechanisms, and production data systems through vision software to achieve a complete inspection closed loop from "seeing a defect" to "automatic judgement, automatic rejection, and data traceability".

This Category Solution

Food industry sub-solutions and related subpages

The food industry can be further broken down into separate solution pages by category and inspection stage. The table below lists the related pages that are already built and accessible; the remaining sub-directions are planned according to the ten product directions above and will go live as their content is completed.

This Category Document Checklist

Confirmed items and outstanding items

Common Question

Common questions about visual inspection in the food industry

Submit sample testing

Send us the product, defect samples, inspection requirements and production line cycle time; a solution engineer will determine which category of inspection problem it belongs to and give recommendations on the imaging method, algorithm route and configuration.