# ML idnetification from Image Data

Identification of Pseudomonas aeruginosa strains. 

Uses data augmentation and transfer learning to overcome data starvation problem. Also uses DL and CNN.

Motivation: colonies, characterization in sizes, shapes, edges, textures and degrees of opacity and color. Categorization. From morphology -> genome sequencing technology. 

Paper: a collection of 69 clinical and environmental P.aeruginosa isolates which present a range of phenotypic features. From these 69 isolates, generate a library of 266 P.aeruginosa images.

D-CNN and data augmentation and transfer learning to classify strains of P.aeruginosa and achieved an accuracy of 94%.

**Morphological variation across strains and replicates**

Isolates -> LB agar plates with Congo red -> quadruplicate and generate images. 

<figure><img src="/files/QjXIpRz51HHb4MUkoSNM" alt=""><figcaption><p>Data</p></figcaption></figure>

The above figure summarizes some of the metrics (B) we use for colony complexity, especially compression ratio and sobel operator. (both in form of vector)

Fig 4- extent of variation using coefficients of variation (CV = mean / standard deviation) across replicates and strains.&#x20;

**classifying from image data**

D-CNN. Use VGG19 DCNN on ImageNet with custom top layers on colony dataset. Apply data augmentation increase 266 images to 38304 images. 74-26 train test split.

First approach, train custom layers with learning rate of $$10^{-3}$$. Second approach is train all layers with learning rate of $$10^{-5}$$.Metrics: accuracy and loss, and use cross-entropy loss. Second approach produced the best performance across all metrics.

Note: downsampled the image to allow the use of pre-trained imagenet. So downsampling could be very important in this type of transfer learning (or, to make sure the image's dimension matches).


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://tianyi0216.gitbook.io/blog/reading-notes/ml-readings/ml-idnetification-from-image-data.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.