Success is a lousy teacher, failure is not.

Success is a lousy teacher, failure is not.

You may know that I have been working on an app to count annual rings on trees using your mobile. You may also have noticed that it has been rather quiet about this for a couple of weeks.

The reason is simple.

I was out testing it in reality and realized that it would not be possible to build such an app. I will now tell you why, so that if someone else tries this in the future, they can build on my experience. 

When you are surfing the web, almost everyone talks about their successful ventures, but few tell about when they fail. It makes you think that it is easy to succeed, but it also stops you from learning from other peoples mistakes. Success is a lousy teacher – failure is not. I am now going to tell you about what I learned and why I am now abandoning the idea.

It started several years ago, I heard a recurring problem that foresters thought it was cumbersome to count annual rings. They preferred a machine that could scan the age of the trees just by holding it up against the tree.

It is a difficult technical challenge to be able to look inside the tree without felling it. It felt beyond my ability. But it maybe would be possible to photograph drill cores with a telephone and then let the phone count the annual rings itself.

 I started building an app to be able to photograph drill cores in a first step and count them manually. If successful, the next step would be to build an AI solution that counts them automatically.

The first challenge was to get close enough, but this could be solved by taking several photographs and then stitching them automatically. It was a bit difficult because the autofocus had a hard time focusing on a narrow drill core. But I solved it by doing a magnification function that showed if the drill core was sharp or not. If the user held a hand behind the drill core, it all went well.

The second challenge was to stitch the images – the user had to have a fair amount of overlap between the images. Not too little, because then it did not manage to stitch and not too much because then the user would have to take too many pictures. I solved it by showing a fair amount of the previous image and giving the user feedback if it managed to stitch the new image with the old image or if it needed to be redone. By using haptic feedback, it became clear whether one succeeded or failed.

I tested indoors and felt I had something going on.

Then I went outdoords a few times and learned the following:

1. The light was a challenge, sunlight and snow created much contrast in the image.

2.The annual rings were not clearly visible in the pictures. It was hard to be really sure if it was spring wood or summer wood. See image 1 and 2.

 
Image 1 Drill core from Pine. It is hard to count the annual rings .

3. There were sometimes bark residues on the drill core. They appeared as brown spots of dirt. When you count manually, you can look from the side of the drill core to see the annual rings that were not covered with dirt.

Image 2. Drill core from Pine. You often get dirt on the drill core. It sometimes makes it harder to count and will make automatic counting even harder.

The problem with the light, could be solved if you made sure to have the sun in the back and have as little snow in the picture as possible.

I did not solve the problem with the annual rings not being good enough. The resolution was quite high, 20-40 pixels per mm. But the sharpness was not razor sharp and the contrast between the annual rings too low. I tried to increase the contrast and apply different amounts of sharpness in the image. But it still did not turn out well enough. This really felt like a crucial problem. The problem mostly occurred in the outer part of the image and it could partly be an optical issue.

The desktop annual ring software that are available on the market often require the analysis of cut wood that are polished. Drill cores are not as good a material to use, and that is probably the main reason why it was so difficult.

The desktop annual ring software that are available on the market often require the analysis of cut wood that are polished. Drill cores are not as good a material to use, and that is probably the main reason why it was so difficult.

Avoiding bark residues on the drill core is not easy, it usually happens for pines that have thick bark. Peeling off this was difficult and you risked losing the drill core while doing so. In addition, it would be a challenge to let an AI distinguish between the dark summer wood and dirt.

What made me finally give up was when I was standing out in the woods, struggling to take overlapping pictures. When I finally succeeded, I had a hard time seeing the annual rings. I switched to counting them manually and it was faster. If I have problems with my own app, then what should the experience not be like for another user?

Here is a movie when I use the app out in the woods. Check it out if you want make your own opinion. I partly regret the attempt to build this app as it did not become a product. But I learned a lot , including the following:

 1. There is always a risk associated when you create new products. You must try to minimize the risk by testing the critical parts first. In this case, it was the photo quality. 

2. Interest has been lower than expected. Few people have shown interest. This indicates that the market for such an app is not large. If you build a great app on a small market, it is hard to keep it viable.

3. Don´t hesitate to give up even though you have invested a lot of time in a project. It is better to test a lot of ideas and choose the one that get most success and traction. Now I have more time to spend on my other ideas….