Findings show AI can provide accurate predictions of mango dry matter content using near-infrared spectral data 

A recent CQUniversity Master of Research graduate has investigated how AI and deep learning models can be used alongside near-infrared spectroscopy to predict mango dry matter content, a key indicator of fruit maturity and eating quality. 

Ramesh Kumar Chaudhary

Ramesh Kumar Chaudhary

Image: CQUniversity

Ramesh Kumar Chaudhary said he was drawn to the project because it allowed him to combine his passion for AI with a real-world agricultural challenge. 

“The opportunity to work on mango quality assessment was particularly interesting because it allowed me to combine my computer science background with agricultural research,” Chaudhary said. 

“I was excited to explore how AI could help improve fruit quality assessment and contribute to Australia’s agricultural industry.” 

Traditional methods of assessing internal fruit quality can require fruit to be cut open and analysed in a laboratory, a process that can be time-consuming, labour-intensive and destructive. 

Near-infrared spectroscopy offers a non-destructive alternative by using light to collect information about the internal properties of fruit. Chaudhary’s research examined how AI models could use this information to provide more accurate predictions of mango dry matter content under varying conditions. 

The study analysed data collected across seven growing seasons, multiple orchards in Queensland and the Northern Territory, and 31 measuring instruments, allowing researchers to test how well AI models performed under real-world conditions. 

Chaudhary said one of the most significant findings was that AI could provide accurate predictions of mango dry matter content using near-infrared spectral data. 

“Artificial neural networks were particularly promising because they demonstrated strong predictive performance and good robustness across varying conditions,” he said. 

The research also challenged assumptions that increasingly complex AI models always produce better results. 

 “One important lesson from the research was that more complex AI models are not necessarily better,” Chaudhary said. 

“Selecting an appropriate model and ensuring that it performs reliably across different datasets can be more valuable than simply increasing its complexity.” 

The findings could help support improved decision-making around harvesting, grading and quality management within Australia’s mango industry. 

“More reliable prediction systems could help growers and packing facilities make better-informed decisions and improve consistency in the quality of mangoes reaching consumers,” Chaudhary said. 

The project was supervised by Dr Arjun Neupane, Professor Kerry Walsh and Dr Zhenglin Wang, whose expertise in artificial intelligence, engineering and agricultural science supported the interdisciplinary research. 

“My research demonstrates how combining AI with non-destructive sensing technologies can contribute to more reliable and efficient agricultural quality assessment,” Chaudhary said. 

“I hope this research contributes to the continued development of intelligent technologies that support Australian growers, improve fruit quality management and ultimately benefit consumers.”