With the use of artificial intelligence (AI), a group of Pakistani researchers has developed a visual categorization approach that reliably estimates the sweetness of locally grown citrus fruits.
The team, headed by Dr. Ayesha Zeb of the National Centre of Robotics and Automation at the National University of Sciences and Technology (NUST), was able to accurately anticipate fruit sweetness with an accuracy of over 80% without causing any harm to the fruit.
The scientists chose 92 citrus fruits from a farm near Chakwal for their experiment, including Blood Red, Mosambi, and Succari kinds. Using a portable spectrometer, they collected spectra, or patterns made from reflected light, from specific spots on the fruit’s skin.
The group used near-infrared (NIR) spectroscopy, a method for analysing non-visible light spectra, to look at the fruit samples. Sixty-four of the fruits were used to calibrate the spectrometer, while the remaining 28 were used to make predictions.
While NIR spectroscopy has previously been used for damage-free fruit classification, the Pakistani team used a fresh approach by modelling the sweetness of regional fruits using this technique. They also incorporated AI algorithms for direct orange sweetness classification, which increased precision.
Chemical and sensory testing have long been the standard methods for determining the sweetness of fruit. Total sugars, or Brix, are used to quantify how sweet an orange is, while titratable acidity (TA) provides an indication of how much citric acid is present. The scientists developed the AI model by peeling off samples from the indicated areas used for spectroscopic and comparing them to reference values for Brix, TA, and fruit sweetness.
The extracted juice was tested for Brix and TA in a laboratory to get accurate readings. In addition, the fruits were eaten by human volunteers who classified them as flat, sweet, or very sweet.
The scientists have improved algorithms for AI
The group trained the AI algorithm on 128 samples using the collected spectrum, reference values, and sweetness labels. The artificial intelligence model was developed to make Brix, TA, and sweetness predictions from spectral data. The researchers used information about 48 new fruits to test the algorithm, comparing anticipated values with those derived from sensory evaluations and chemical analysis.
Amazingly, the AI model not only performed as well as or better than conventional approaches in predicting sweetness, but it also precisely predicted Brix, TA, and total sweetness values. The model was 81.3% accurate overall when testing for the detection of sweet, sour, and neutral flavours.
The application of this scientific finding to the citrus business, and more specifically to the evaluation of citrus fruit quality, is profound. Oranges, unlike bananas and mangoes, do not continue to ripen after being picked. As a result, the industry might benefit from this cutting-edge AI-based approach to evaluating the sweetness of citrus fruits, and consumers would be more satisfied.
Since Pakistan is the world’s sixth-largest citrus fruit producer (with exports of 0.46 million tonnes in 2020), it stands to benefit from this development. Nature, a highly respected scientific publication, has published the results of this study.
Dr. Ayesha Zeb and Dr. Mohsin Islam Tiwana of NUST’s National Centre of Robotics and Automation led the project’s multidisciplinary team of researchers, which also included Dr. Waqar Shahid Qureshi of the School of Computer Science at Technological University Dublin in Ireland; Drs. Abdul Ghafoor, Muhammad Imran, and Alina Mirza of NUST’s Military College of Signals; Dr. Amanullah Malik of NUST’s department were also the honorable participants of the study. Studies like like this will pave towards the road for innovation and scientific achievements and achievements,
To read our article about “New high tech 22 bed cardiac unit set in Liaquat Hospital” click here
