Interesting computer vision articles

[jan-kleks]

I have recently stumbled upon this very interesting computer vision article about recognizing species of insects which can be hard to identify. Maybe you will find there some ideas for further improving the iNaturalist project.

Automated Taxonomic Identification of Insects with Expert-Level Accuracy Using Effective Feature Transfer from Convolutional Networks -- https://academic.oup.com/sysbio/article/68/6/876/5368535

Keep up the great work!

[jan-kleks]

Other interesting articles:

• How we accidentally built the most accurate AI insect detection app -- https://www.linkedin.com/pulse/how-we-accidentally-built-most-accurate-ai-insect-detection-krish

🤔 Does this article imply that it might be enough to remove background/shadows (possibly this stage could be automatized with another algorithm/AI) to get much better training results overall?

• Classifying the Unknown: Identification of Insects by Deep Zero-shot Bayesian Learning -- https://www.researchgate.net/publication/356706523_Classifying_the_Unknown_Identification_of_Insects_by_Deep_Zero-shot_Bayesian_Learning

• The Benefits of Using Stable Diffusion AI in Image Recognition: https://ts2.space/en/the-benefits-of-using-stable-diffusion-ai-in-image-recognition/

• The Power of AI-Generated Training Data: Image Recognition With DALL-E

• TomatoDIFF: On-plant Tomato Segmentation with Denoising Diffusion Models

• Accurate detection and identification of insects from camera trap images with deep learning

• Textural Analysis for Medicinal Plants Identification Using Log Gabor Filters

• Combining computer vision and deep learning to classify varieties of Prunus dulcis for the nursery plant industry

• Image-based taxonomic classification of bulk insect biodiversity samples using deep learning and domain adaptation

• Machine learning for expert-level image-based identification of very similar species in the hyperdiverse plant bug family Miridae (Hemiptera: Heteroptera)

[LDNOOBWV2]

I read the articles you linked here and I found another one that might be interesting, using "Deformable Convolutions": https://arxiv.org/pdf/1703.06211
About the Log-Gabor-Filters: One might try with simpler filters in the data pipeline like:

• "Histogram Equalization" and/or "Gaussian Blur" (Example: https://www.geeksforgeeks.org/ai-in-image-processing/) and/or a
• "Canny Edge Detector" and/or
• "Sobel Derivates" and more (OpenCV Tutorial: https://docs.opencv.org/3.4/d7/da8/tutorial_table_of_content_imgproc.html)
  Another one from the Pillow Framework is:
• ImageFilter.EDGE_ENHANCE (https://pythonexamples.org/python-pillow-image-edge-enhance/)

In one AI project for image classification (pharmaceuticalIndustry, drug plants identification) we successfully used:

• "Histogram Equalization" and
• "Gaussian Blur" and
• "Pillow Edge Enhance" in the pipeline. We tested "Canny Edge", "Sobel" and "Scharr" as well, but the above mentioned combination worked best.

In another project for object-detection (Self Dirving Cars) we tested "Deformable Convolutions" vs. "Dilated Convolutions" and decided due to better results to use "Deformable Convolutions".
Hope some of the ideas help.