How to turn a fish-processing facility into a super-efficient enterprise

The customer for this case is a full-cycle fish-processing facility: from incoming fish to finished canned fish, seafood cocktails, and fish snacks.

The company opened a tender to find a contractor in the field of neural network analytics and machine vision for two of its separate branches. In doing so, each branch solved its own problems and made an independent decision to select a future contractor. As a result, one branch, headed by the risk management department, chose our company, while the other (headed by the production manager) went with a different company. Looking ahead, it should be noted that our neural network analytics had already been implemented at six sites in four production buildings of the company, while the project of the second branch had not started yet (it had been a year!).

Communication with a new customer always starts with our question: "What problem do you want to solve with video analytics?" Then we focus the question on this problem and ask to show a video or photo report to understand the specific business process. In this case, our experience allows us to determine already at the first meeting not only how we can solve the problem, but also to orient on the price.

Integration is the process between the signing of the contract and the transfer of the finished solution to the client.

In this case, the training of neural network robots, setting up a personal account for the customer (in which you can monitor the work of your company from anywhere in the world) takes 3-4 times less time than that for the issues of coordination: specifications, equipment, places of its installation, the accuracy of control metrics and the contract. And the duration of this period is due to the approval procedures adopted at the customer's enterprise.

The implementation process itself took less than two months and consisted of several stages.

Survey stage. At this stage we determine the feasibility of using the company's current video surveillance system (VSS) to implement neural network video analytics. The task of our engineers is to make a simple conclusion - whether we can use the existing VSS, whether it needs to be changed or partially updated (for example, the recorder is suitable, but the camera needs to be replaced). The client receives a specification with a list of necessary equipment, if upgrading is still required. But in most cases we use a system that is already available in the enterprise.

In this case, no modernization was required, we used the customer's existing equipment.

Robot training phase. The stage of training the robot, which interacts with the video system of the object, identifying what is happening in the video. After that, the robot begins to "understand" what is in the video and what is not, on the basis of which the relevant reports are generated. As a result, the customer is conveniently informed about all violations with video confirmation of each event.

As for the specifics of the industry, we should note the presence of the so-called demountable lines. We first encountered the assembling and disassembling of lines in fish production.

The point is that the production is equipped with a large number of light conveyor lines, which, about once a month, are completely disassembled for washing and then reassembled. Accordingly, the first assembly after washing and ... all the equipment was not assembled where it was originally (a meter to the right or left or changed the configuration). Despite the fact that we discussed this point at the start of the integration, but in practice the situation still occurred. The line was re-assembled, and then we put the key marks on the floor, which determined how the lines should be assembled after each wash.

There are other nuances, such as shops with high humidity, with a lot of steam and condensation, which affects the placement of CCTV cameras. Each production has its own nuances, but they are all solved.

This is now a standard case for us, and subsequent implementations at three fish processing companies have gone even faster. But each new area is incredibly interesting for us, especially if the client is ready to create production trends in his area.

One of the principles of our work is to reduce the cost of video analytics implementation as much as possible. The goal is to make the project as passable as possible for the company from the point of view of financial efficiency. And in most cases such answers are found together with the customer. For example, if the customer has 10 tasks, 9 of which make up 10% of the budget, and another 90%, we try to decide together how to make this one cheaper, or the customer can refuse from it. In the joint work it is important to clarify which issues will give the best result.

Modesty is said to be more trustworthy. We believe that it is the facts and the results that are trustworthy.