Jorrit van Gils

Content:

  • About

  • Projects

  • CV

  • Teaching

  • State Forestry youth board

About

I’m a final-year Msc student Forest Nature conservation at Wageningen University and Research, advised by dr. Gert Kootstra (Computer vision) and dr.ir Patrick Jansen (Ecology). Experienced Python and R programmer with a demonstrated history of working as teacher biology. Internationally oriented with a strong interest in technology-based solutions for conservation. Skilled in machine- and deep learning models, pose estimation, GIS-applications and networking. Volunteer at state Forestry, chair of its communication work group. Responding to how young people experience nature.

Projects

Thesis
Personal interest and various programming courses have led to a broad knowledge in the field of programming and artificial intelligence (AI). This thesis offered me the opportunity to apply this knowledge to wildlife camera images. These courses where:

  • Geo Scripting
  • Programming in Python
  • Machine learning
  • Deep learning

In my thesis I work on predicting Red deer (Cervus elaphus) behaviour from wildlife camera trap images by comparing two methods. The first method classifies behaviour with a YOLO network directly from Red deer images. The second method does not classify directly from images but classifies based on the position of predicted animal key points that are derrived from a pose estimation method DeepLabCut. The key points representing the animals pose are used as input for a MLP 1D CNN network. I’m particularly interested in the the benefits of a key point-method compared to a behaviour classification method that classifies behaviour directly from images. Potential advantages of using pose estimation might be that irrelevant contextual information is left out and that less data is required to train a rubust behaviour classification model. Challenges in AI-methods and insights in animal behaviour are a great motivation for me to develop myself further in this field in the future. Personal suggestions for future research can be found in this document.

(Source: BOX21 and DeepLabCut)

The first step in my thesis was to filter from 454904 animal image sequences to 506 usable Red deer images. Then they where labeled with behaviour. Interest in the R script from other students resulted in a workshop (5-11-2021) in which I taught the students how to select images and annotate behaviour for their own purpose. Finally, I explained how to add relevant columns like the habitat type, the URL and identifiers.

Click here for the behaviour annotation script in R or view the final table:

Machine- and Deep learning courses

Concepts covered machine learning course:
Linear regression, logistic regression, K-nearest neighbours, LDA, QDA, Cross validation, bootstrap, SVM, kernels, trees, boosting, bagging, random forest, PCA, hirarchical clustering, K-means, Gaussian distribution

Concepts covered deep learning course:
Linear regression, stochastic gradient descent, softmax regression, multi-layer perceptron, backpropagation, regularization, CNN, Data augmentation, computer vision, semantic segmentation, transposed convolution, fully convolutional networks, RNN, GRU, encoder-decoder, Attention, LSTM

Deep learning project
At the end of the deep learning course where two weeks of project. My task was to build in a deep learning model in Python from scratch that could recognize land use type from UMC satellite images.

Click Here for my Google Collab deep learning Python script

alt text alt text
(Source: Google Collab)

Internship

My internship was performed at State Forestry (Staatsbosbeheer) and included a brook system analysis in the provice of Gelderland (The Netherlands). I mapped different aspects of the Soerensche Beek, like the waterquality, soil composition and historical maps. Extra courses gave me a good understanding of how satelite images comes about and how these images can be used in practice.

Some examples of maps that I created:

Supervisor: H.Sluiter, ecologist State Forestry Grade: 8.2
Feedback: Jorrit is eager to learn and very curious. In a short time he has developed a profound knowledge of the Soerense brook system. Therefore he has studied new scientific areas such as GIS, water quality. He is also keen on doing as much as possible to broaden his knowledge. Outstanding is his way of communicating and able to explain complicated matters. Not only did he build up a network of external contacts. He was also very conscientious in the communication with its supervisor (me). The report is clearly written and leads to a number of important conclusions that, substantiated in this way, will also be used in the WFD. Jorrit has all the qualifications to work for one of the Nature Conservation Organisations.

Mathematics

  • Mathematics 1
  • Mathematics 2
  • Practical Aquatic Ecology and Water Quality
    • data analysis and mathematical modelling

Statistics

  • Statistics 1
  • Statistics 2
  • Ecological methods 1
    • univariate & multivariate statistical techniques
    • Excel, R, Canoco
  • Wildlife ecology and conservation
    • Zonation, Maxent

Datacamp and Coursera courses

  • Introduction to R
  • Intermediate R
  • Introduction to Importing Data in R
  • Introduction to the tidyverse
  • Intermediate Python
  • Deep Neural Networks with Pytorch

CV

Download my CV here

Teaching


From 2016 until 2019 I was working as biology teacher and mentor on high school De Nassau in Breda. Here I taught students in their first 3 years of the lower- and higher general secondary education. I developed personal skills in communication and in organisation. I designed my own digital biology modules with prowise software allowing interactions with students tablets. Despite the big learning curve, I decided to quit my job and further develop in data science, AI and conservation by doing a master Forest Nature conservation at Wageningen University.

Volunteer


I’m member of the Dutch State Forestry youth board. We advise the national company solicited and unsolicited and give insight in how youth are experiencing nature. Since I have been in the board for 2 years, my additional role is chairing the communication workgroup. Part of this role includes managing our social media platforms Instagram, Linkedin and Facebook. Click here for more info (page in Dutch)