The military-industrial scientific research system of the academic master.

It is naturally impossible for the delegation from Snecma to wait in Shengjing for ten days.

As for the idea of ​​further cooperation, although there is one, it is impossible to make a decision on the spot.

Therefore, after agreeing on some more details the next day, Ian Vista and other core members of the technical team returned to France.

Only a few people were left behind, along with Bernard Sean and others, waiting for the Chinese side to officially deliver the batch of samples for testing.

Liu Yongquan reported the meeting situation and new production requirements to the group as quickly as possible.

The technical team of the Haojing Institute of Optics and Fine Mechanics has successfully assembled the first ultra-short laser processing equipment for mass production at the Liming Factory, but it is still in the process of debugging and testing and will not be able to go into actual production before the end of the year.

Therefore, at least for now, this batch of M88-3 parts still have to be handed over to the machine in the Torch Laboratory for hole making operations.

However, since the technical development stage has been basically completed, including the fact that even the parts of the M88-3 are not produced for the first time, Chang Haonan only needs to take charge of it.

The specific work at the software and hardware levels can be completed by Li Yabo and Academician Hou.

Of course, this doesn’t mean that Chang Haonan deliberately wants to be a hands-off boss.

Although he is no longer very interested in the M88-3 project, the SeA series engine, as the first medium-to-high bypass ratio engine to be used by the Aviation Industry Corporation of China, is still worthy of attention.

In fact, Chang Haonan even planned to use the development of SeA650 as a training exercise before the official launch of the WS20 project, so he specially recruited many members from the Gaofa Group to participate.

The reason why he had no time to take care of the situation at AECC for a while was that after he came back from Gaojing last Friday after seeing Tu-160, he only had time to rest for one day before he received a call from the China Aerospace Science and Technology Corporation.

Invite him to a project development meeting.

As for the content of the meeting, it was naturally about the onboard software of the Ocean No. 1 satellite.

It was also a coincidence that Chang Haonan had some new ideas on target contour recognition while developing new algorithms for the TORCH Multiphysics software a few months ago.

Therefore, he spent some time preparing for this matter seriously.

And it was almost at the same time that Vista and others boarded the plane back to France.

Chang Haonan was in his own laboratory, giving a lecture to a group of technicians from the Space Technology Research Institute, also known as the Fifth Academy of Aerospace...

lecture.

That's right.

Similar to most of the projects he has been involved in in the past.

During the project development meeting, Chang Haonan introduced his contour target extraction model in the shortest possible length.

But in an environment like a conference room, it was obvious that the effect was not good enough for him to speak alone.

Besides, the conference also included experts in the pure hardware field and even rocket technology that had nothing to do with satellites, so he couldn't drag everyone together to listen to his long speech.

Therefore, after realizing the problem, Chang Haonan immediately chose to quit while he was ahead.

Then we held a small meeting with some people related to software development and software and hardware adaptation.

"Mr. Chang, the parametric active contour model (snake model) you mentioned yesterday was also systematically demonstrated by our institute two years ago."

The speaker was a young technician named Lin Shenming who was sitting at the first seat on the left side of the conference table:

"But in the end, we found that although the snake model can provide smooth and closed edge contours that meet pixel requirements, it cannot adaptively control changes in the curve topology. It is also difficult to clearly segment the concave areas of the target object, which is a big problem for ship... uh... marine feature target recognition."

Although he is only in his thirties, Lin Shenming is actually the deputy chief designer of the ground application system of the Ocean No. 1 satellite.

If Chang Haonan had not been standing at the forefront at this moment, he would obviously be regarded as one of the most dazzling young talents in the entire Ocean No. 1 project.

Of course, his current problem does not involve any personal emotions.

I just have some concerns about the active contour model technology route.

Although Ocean No. 400 has solved the hardware problem and can even be said to have obtained the world's top optical system, it is active in a low-Earth orbit at an altitude of kilometers after all, and the photos it takes are still blurry after being enlarged.

Therefore, it is extremely important to accurately identify specific contours from a pile of vague and chaotic targets.

The prerequisite for recognizing contours is to separate the contour shape from a blue or black background.

If Ocean No. 180 is just a regular ocean resource satellite as its name suggests, then the inability to split the concave area, that is, the area with an internal angle greater than °, is not a big deal.

But as a de facto ocean surveillance verification star, this problem is completely unacceptable.

It is true that aircraft carriers are large.

But among the ranks of civilian ships, there is no shortage of behemoths whose sizes are close to or even exceed those of aircraft carriers.

The biggest feature of an aircraft carrier is its full-length deck, which, when viewed from satellite perspective, is similar to the wide front deck of some civilian transport ships.

Even a container ship fully stacked with containers can be easily misjudged.

This leaves only two features that can be easily exploited.

First, head towards the island on one side.

Apart from warships such as aircraft carriers and amphibious assault ships that have the need to carry aircraft, no other ships would use this bizarre design.

Civilian ships will place the wheelhouse as close to the front or rear end of the hull as possible.

But the problem is that the island is indeed a bit too small.

Especially for nuclear-powered aircraft carriers, since there is no need for a smoke duct, the island can be made very small and far back.

And it also overlaps with the deck part to a great extent.

Whether it is radar, visible light or infrared light, it is difficult to identify if the resolution is insufficient.

In comparison, another feature, the irregular hull outline formed by the angled deck and take-off deck, is more friendly.

Due to considerations of volume ratio, it is impossible to design a civilian ship with such a jagged appearance.

Even small and medium-sized aircraft carriers without angled decks and amphibious assault ships can be excluded to a certain extent.

What’s more important is that the 300-meter-long and 40-meter-wide outline can still be barely segmented with the current satellite resolution.

Therefore, it is necessary to solve the problem of recognition rate of concave areas.

Otherwise, if the angled deck is divided into smooth arcs, it will look like an oil tanker without any distinctive features.

"Mr. Lin is right."

Chang Haonan walked to the blackboard:

"It is difficult to directly apply several traditional active contour models to our Ocean No. 1 image recognition algorithm."

"But when I was writing a paper before, I accidentally saw an idea, which is to use the variational level set method to improve the active contour model."

"So far, the main products of this line of thought are the geometric active contour model and, further, the geodesic active contour model developed by segmenting the variational level set method using the Mumford-Shah functional."

"Of course, since level set methods have been subject to non-conservation issues in the past, the adaptability of these two methods in the face of changes in the curve topology is still limited."

"However, if you have read my previous article published in the first issue of JCAS, you will know that I have solved this problem from a theoretical level..."

After that, Chang Haonan wrote the first equation on the blackboard:

p(F)=e^(-βE|F|)/Z.

At this moment, another person raised his hand:

"Mr. Chang, I have read most of the text in the first issue of JCAS, but if I remember correctly, the article you published seems to be used for multiphase flow simulation?"

"Yes."

Chang Haonan just finished writing the equation at this time, so he turned around and replied:

"But the mathematical principles remain essentially the same no matter how they change."

“The difficulty of multiphase flow simulation lies in the fact that the topological structure of its phase interface is highly uncertain, so the moving interface needs to be described as a level set that changes over time.”

"Thus, this method can be applied to the evolution of curves in partial differential equations with almost no modifications."

After he finished speaking, he pointed back to the blackboard:

"This is the Bayesian form of the Gibbs formula in statistical mechanics, and the variational method I just mentioned happens to be completely consistent with it in form..."

"The image processing method based on partial differential equations is essentially based on the continuous mathematical model of the image. It assumes that the image changes according to a specified partial differential equation, and the solution of the PDE is the desired processing result."

"As for how to determine the partial differential equation form of this formulation..."

As Chang Haonan spoke, he drew an arrow between "image change" and "partial differential equation":

"Generally speaking, the desired image changes are compared with some mathematical and physical processes, such as comparing image smoothing to the diffusion process of impurities. Of course, most processing methods are not so simple, which is exactly what we need to study at this stage."

"Based on my current research progress, as long as I can get to this point."

He circled the words “partial differential equations”:

"The subsequent numerical solution process is almost not a problem!"