Grading & Sorting Solutions – Technology Analysis

November, 28, 2016

Ellips Overview

Ellips is a 40 person organization that is solely dedicated to the development and continual enhancement of our grading and sorting technology.  This dedication has resulted in industry leading solutions and our ongoing commitment to R & D ensures that our customers will benefit from the continual improvement of our technology.

Historically, Ellips has been on the forefront of technological advancement including the development of our own operating system (HEROS) that significantly reduced the overhead associated with commercial operating systems such as Linux.  By increasing processor utilization, we are able to take better advantage of new technologies including higher resolution cameras.  As a result, Ellips was able to incorporate HD cameras and LED lighting into our grading solutions 3 to 4 years prior to most of our major competitors.  We have hundreds of HD/LED systems in use and this experience has enabled us to become the industry leader as it relates to optics, lighting, camera placement and overall image quality and consistency.

Ellips works closely with all of the enabling technology providers (i.e., camera, lighting, computer manufacturers, etc.) to ensure that our solutions incorporate the best technology available.  In fact, our R&D staff conducts independent analysis of these components specifically as it relates to their use in our environments and applications.  As an example, we recently identified a deficiency inherent in the spectrometers used for our Internal Quality solution.  By working with the spectrometer manufacturer (the industry leader), we convinced them to make a modification that enhanced the performance of our IQ solution.

Ellips strives to achieve the optimal balance between inputs (i.e., images, spectra), processing power, software capabilities (i.e., grading algorithms) and system complexity in order to deliver superior performance.  If any of these elements are out of balance, system capabilities and, ultimately, your grading performance will be negatively impacted.  In other words, technological advancements are meaningless if images overrun processing power, complexity negates usability, unreliability impacts uptime and excessive costs outweigh potential benefits.  It is somewhat analogous to a packing line where increased capacity in one area (i.e., in-feed) would provide little benefit if the other components (i.e., sizer, packing or box handling) can’t handle the additional volume.

It is important to note that the performance advantage delivered by Ellips has been achieved over time with the benefit of experience and customer feedback.  Short cuts are tough to come by.  You can be assured that we are continually working to improve our solutions through more efficient designs, enhanced software and the use of new technology.  These efforts help us deliver ongoing value to our customers while maintaining competitive differentiation.

 More Is Not Always Better

Recently, some interesting statements have been made regarding the capabilities of new optical grading & sorting solutions.  For example, there have been claims made about the generation and utilization of “up to” 500 images per piece of produce.  It is important to note, that not all images are equal and images are only part of the story.    At this time, virtually all commercially available computers that combine high level performance with reliability, supportability and “reasonable” costs utilize the Intel Core I7 processor.  Although performance can vary depending on operating system and related “overhead”, computers equipped with these chips can typically process about 500 MB/sec.  A single HD (1280×960) camera taking 100 images per second (10 cups/sec with 10 images/apple) will generate 62.5 MB of data per second.  Assuming 4 HD cameras per lane (400 images/sec/lane), 250 MB of data would be generated per lane per second which means one computer would be required per 2 lanes.  Full HD+ resolution (1920×1280) would generate twice the data thus requiring one computer per lane.

Taking 500 equivalent images (i.e., HD or HD+) of each apple, at a sizer speed of 10 cups/sec, would generate 5,000 images per sec/lane.  Processing this quantity of images would require roughly 6 computers per lane for HD quality or twelve computers per lane for HD+.  Obviously, this scenario is currently unrealistic based on cost, complexity, footprint, as well as, power consumption and cooling requirements.

By deploying 8 cameras (HD not HD+) per lane, you could theoretically process 80 images for each apple with 1 computer per lane.  However, our testing associated with additional cameras did not reveal significant advantages in defect detection especially, when considering the added complexity and cost.  Since we already see virtually all of the apple, a high % of the time, Ellips’ main focus is on improving its software to more accurately assess a potential defect (i.e., deep stem bowl crack vs. the actual stem) vs. utilizing more pictures of the same defect.

Based on the calculations above, it is apparent that the definition of “images” can vary by manufacturer and could include HD and non HD pictures, sensor input, individual wavelength measurements, etc.  In reality, the quantity of images can be quite misleading as it relates to actual grading performance.  We have recently seen a manufacturer “scale back” some of their images to 640×480 in order to process more images.  Below is an example of these images (notice the poor resolution) from a competitor’s system that also incorporates a “block” overlay approach (vs. an overlay depicting the actual shape of the defect) in order to further reduce processing requirements.  Compare the first set of images to those that follow which were generated from an Ellips system.  You will also notice that Ellips tracks the same defect across multiple images to better determine individual defect characteristics and to avoid “false positives”.



The above analysis reconfirms that more is not always better.  Increasing the number of images without the ability to properly utilize them will add cost and complexity without significantly increasing performance.

Not All Solutions (IQ) Are Created Equal

Ellips IQ is a proven solution that is used to assess the internal quality of various commodities, including apples, onions and pomegranates.   Our approach which utilizes transmittance technology (light passing through an object) has demonstrated far superior performance compared to competing solutions.  Based on technology experience and comparative testing, it is our opinion that most other solutions will need to be significantly improved or redesigned in order to match our current capabilities.

That doesn’t mean that other systems can’t detect internal quality issues.  Rather, it implies that under certain conditions, inferior performance would render a system unusable.  For instance, a system’s lack of detection accuracy might not bring your packed product into “grade”.  Alternatively, you may be able to achieve grade, however, to do so, the % of good fruit being rejected with the bad is unacceptably high.

For example, an experienced user of a competitive solution typically found that 60% of their rejected apples were actually good when they were trying to achieve 2% in-grade goal.  Imagine 10,000 bins of premium fruit (i.e., Juici, Honeycrisp, Envy, Jazz, etc.) with 10% (equivalent of 1,000 bins) of the apples affected by internal issues.  In the process of detecting the 1,000 bins of bad apples, the competitive solution would incorrectly reject 1,500 bins of good apples or the equivalent of 30,000 boxes of packable fruit.   The Ellips IQ solution would achieve the desired QA result while incorrectly rejecting much less good fruit (i.e., 250 bins or less).  The difference of incorrectly rejected bins (1,250) would equate to roughly 25,000 boxes of packable fruit or $1,500,000 at $60 per box.  Even though this is a rough calculation, it illustrates the potential value of one solution vs. another.

Pin It on Pinterest