Inlägg Currently viewing the category: "image analyzing"



The Beescanning tutorial shows how to take pictures optimal for image analysing. Beescanning_tutorial_1 from Björn Lagerman on Vimeo…(Read More)

Good image quality


Image quality is essential for analysing results. Take image from a distance of about 20 cm. Preferably in the shadow, reducing over and underexposed areas. Use both hands and/or a support to avoid motion blurr…(Read More)

Consider this to take a proper image for analysing. 1. Correct distance about 20 cm 2. Focus. Hold your cellphone steady with both hands. Place the comb on a stand. 3. Avoid direct sunlight. Shadow gives better images without sharp contrasts. 4. MAny bees. A collection from a colony should contain more than 5000 bees…(Read More)

Counting bees


To be able to calulate % varroa the software must beable to calculate both number of bees and bees wit varroa. The used techniques are quit different. In the bee counting task the image is analysed  for “density” We are now in the final image development phase. First image below shows results after 22 500…(Read More)

New application – 2.0


“Monitoring a biological system.” A new application for support of BeeScanning 2.0, the next step. This is for the period 201807-201907. Here we focus on developing technology to asses the bee colony health via several indicators measured by new technique. In our third step we aim to implement our new knowledge into new…(Read More)

Varroa determination


In a seminar today we discussed how to decide if an object is a varroa mite or not. Though in Swedish you may find interesting images. For instance the first image is a bee louse not a mite. And region 2157 further down is not a varroa. What do you think it is? Any suggestions…(Read More)

For the first time ever an app can detect the queen. Its very easy to use and lightning fast. The analyse takes 100 milliseconds and in total just a few seconds depending on your internet connection. We soon will add; bees-with-varroa, bees with deformed wings and several other features…(Read More)

Two major breakthroughs


Two major breakthroughs this week: The testing material used to validate neural network training has for the first time shown correctness 100% in combination with 100% accuracy. We have for the first time received correct and verified proposals of previously , by man , unmarked bees with varroa. We arer commensing desig of protocol and…(Read More)

Two promising results from training today. The neuron net may be described as a baby learning to talk. Very varying, sometimes nothing is detected sometimes all varroa! Very promising progress. (It is the neuronnet that has labeled varroa in the images not a human) Results are especially interesting since the images didnt come from the…(Read More)

Teorem introduces itself


Our partner Teorem AB has updated its site. (Only in Swedish yet) Teorem describes its focus on machine learning and Deep Learning. Teorem has cutting edge technology based on its own research…(Read More)