Automatic reading of postal codes with Deep Learning
“Thanks to automated reading of postal codes, with the increase in the % of reading of postal codes and rotated images, an increase in productivity in the process of reading packages would be generated, reducing manual processing of rejected packages.
This would improve fluidity and reduce times, streamlining the process, which has repercussions throughout the supply chain. This also opens up the possibility of reallocating resources to other higher-value activities.”
A postal codes reading use case
Courier and delivery companies and postal operators undoubtedly play a key role in parcel delivery.
To provide these kind of companies with new tools to automate the process, we tested AllRead in a project consisting of a pilot test aimed at improving accuracy and efficiency in the reading of postal codes for letters and parcels, as thousands of them were being previously rejected during the sorting process and require manual intervention.
In this context, we tested AllRead to apply our intelligent reading technology based on Deep Learning and Computer Vision to improve efficiency in an operational environment.
• Industry: post office
• Use case: postal codes identification and reading
More than 15% error rate
In this operational context, letters and parcels with postal codes were processed through several stages of automatic and manual reading and checking.
Both reading processes had great importance for the operational efficiency. And this was so because a large number of letters and parcels passed through sorting machines every hour. And of these, it is estimated that their traditional reading systems were rejecting more than a 15%.
As a result, they were referred to a video coding process, in which their specialized staff manually entered the code that appeared in the image. This resulted in delays because, since automatic processing was not possible, repetitive manual checking tasks had to be performed.
THE USE CASE
Processing and reading of postal codes on letters and parcels
LITTLE ACCURACY IN PREVIOUS READING SYSTEMS
With previous error rates or inability to read above 15%.
MULTIPLE MANUAL PROCESSES FOR VERIFICATION
With video coding and manual review techniques, with the need for human intervention.
Due to the multitude of revisions and repetitive manual tasks required.
More precision and less repetitive tasks
With this approach, the project consisted of automatically identifying and reading postal codes on letters and parcels using Computer Vision and Deep Learning, differentiating it from the previous process.
In doing so, the project represented an opportunity to improve several aspects:
Increasing accuracy rate
able to read codes that other solutions did not recognize.
Reducing repetitive tasks
allowing operators to work more comfortably, spending less time.
Improving operating efficiency
to be able to allocate these resources to higher value-added tasks.
Thus, we worked on demonstrating the capacity of our solution and improving the current reading process in the Post Office sorting machines, by delivering with maximum certainty and efficiency the postal code that their traditional reading systems did not read (OCR and others).
64% more efficient
At the end of the project, our technology showed very positive results, superior to previous systems. The company had managed to read 64% more of the previously rejected images by other OCR reading systems with total accuracy.
Calculating at least 10 seconds of manual pre-processing time for each image, and extrapolating this figure to the large number of letters and parcels that circulate daily, one can imagine the improvement
that AllRead would bring when implemented in the long term, which would also make it easier for operators to free up some of their time to devote to more value-added tasks.
Moreover, since it is based on Artificial Intelligence, the results will be exponentially better the longer the technology is used and the greater the volume of codes processed.
With this test alone, the results have already had a great impact on the company, saving more than 16 hours in repetitive tasks, which demonstrates the great potential added value.
Outperforming Other Systems
Read 64% more of the images previously rejected by other reading systems with complete accuracy.
Reducing Repetitive Tasks
Saved more than 16 hours of repetitive tasks, also making it easier for operators to free up some of their time for more value-added tasks.
Better future results
Due to Artificial Intelligence, the results would be exponentially superior with each retraining of the neural network. This is the power of our technology.
If you want to know more, schedule a meeting with an expert. We will show you how our technology can improve your company’s operational eficciency.