Abstract:Starches are important energy sources found in plants with many uses in the pharmaceutical industry such as binders, disintegrants, bulking agents in drugs and thus require very careful physicochemical analysis for proper identification and verification which includes microscopy. In this work, we applied artificial intelligence techniques (using transfer learning and deep convolution neural network CNNs to microscopical images obtained from 9 starch samples of different botanical sources. Our approach obtained an accuracy of 61% when the machine learning model was pretrained on microscopic images from MicroNet dataset. However the accuracy jumped to 81% for model pretrained on random day to day images obtained from Imagenet dataset. The model pretrained on the imagenet dataset also showed a better precision, recall and f1 score than that pretrained on the imagenet dataset.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to accurately classify starches from different sources through microscopic images in drug formulations. Specifically, the researchers applied artificial intelligence techniques, especially using transfer learning and deep convolutional neural networks (CNNs), to process microscopic images of starch samples obtained from 9 different plant sources. The aim is to develop an efficient and accurate method to replace the current time - consuming and expertise - required physicochemical analysis methods, thereby avoiding potential risks and cost increases caused by starch sample confusion during the drug formulation process. The key points of the paper include: - **Background**: Starch has multiple uses in the pharmaceutical industry, such as being used as a binder, disintegrant, filler, etc. Therefore, correctly identifying and validating starch samples is crucial for ensuring the quality of drug formulations. - **Method**: The study adopted two pre - trained models, one based on the MicroNet dataset (which contains a large number of microscopic images), and the other based on the ImageNet dataset (which contains daily random images). By adjusting the output layers of these models, the researchers constructed a model containing three fully - connected layers for the starch sample classification task. - **Result**: The experimental results show that the model pre - trained on the ImageNet dataset achieved an accuracy rate of 81% on the validation set, while the accuracy rate of the model pre - trained on the MicroNet dataset was 61%. In addition, the model pre - trained on the ImageNet dataset also performed better in terms of precision, recall, and F1 - score. - **Discussion**: Although the images in the MicroNet dataset are more similar to the microscopic images in the study, the model pre - trained on the ImageNet dataset performs better. This may be because the ImageNet dataset contains more diverse images, which helps the model learn more general features and thus generalize better to new tasks. In conclusion, this study demonstrates the potential of artificial intelligence techniques in improving the efficiency and accuracy of starch classification, providing a new tool for quality control in the drug formulation process.

Application of Artificial Intelligence in the Classification of Microscopical Starch Images for Drug Formulation

Confocal Raman Microspectroscopy Combined with Spectral Screening Algorithms for Quantitative Analysis of Starch in Rice

Nondestructive identification and classification of starch types based on multispectral techniques coupled with chemometrics

Hybrid methods for detection of starch in adulterated turmeric from colour images

Deciphering Microorganisms through Intelligent Image Recognition: Machine Learning and Deep Learning Approaches, Challenges, and Advancements

Microscopic and spectroscopic characterization of rice and corn starch

Deep learning neural networks for acrylamide identification in potato chips using transfer learning approach

Automated Interpretation of Blood Culture Gram Stains by Use of a Deep Convolutional Neural Network

Potato diseases detection and classification using deep learning methods

Non-coding deep learning models for tomato biotic and abiotic stress classification using microscopic images

Application of data-intelligence algorithms for modeling the compaction performance of new pharmaceutical excipients

AIMIC: Deep Learning for Microscopic Image Classification

Using Deep Convolutional Neural Networks to Circumvent Morphological Feature Specification when Classifying Subvisible Protein Aggregates from Micro-Flow Images

Artificial intelligence-driven microalgae autotrophic batch cultivation: A comparative study of machine and deep learning-based image classification models

Plant Disease Diagnosis and Image Classification Using Deep Learning

Testing Precision Limits of Neural Network-Based Quality Control Metrics in High-Throughput Digital Microscopy

A large-scale optical microscopy image dataset of potato tuber for deep learning based plant cell assessment

Ability of visible imaging and machine learning in detection of chickpea flour adulterant in original cinnamon and pepper powders

Grading and fraud detection of saffron via learning-to-augment incorporated Inception-v4 CNN

Detection of fraud in ginger powder using an automatic sorting system based on image processing technique and deep learning

Computer Vision and Machine Learning Analysis of Commercial Rice Grains: A Potential Digital Approach for Consumer Perception Studies