A smartphone app that can identify and classify disease from pictures of chicken faeces has been developed to aid poultry veterinarians and farmers in making accurate and swift diagnoses.
Scientists have tested a system based on deep learning algorithms which detect objects and classify images. The algorithms were trained using a dataset of over 8,000 annotated images of chicken faeces, which were classified as healthy, Salmonella, Newcastle disease or coccidiosis.
They found that it identifies areas of interest with an accuracy of 87.48%, and can then classify them with 98.7% accuracy.
The researchers then developed an app called KUKU, to allow farmers and veterinarians easy access to the tool, where they can photograph chicken droppings and have them assessed for signs of the diseases.
Better assistance on farms
“Common poultry disease detection methods include observing the behaviour, physical appearance, type of droppings of the birds, and laboratory examination of sample of chicken’s droppings. Some of these methods, however, are prone to human error, while others are difficult to implement on a regular basis,” they explained in the journal Smart Agricultural Technology.
The new approach may prove more effective than those suggested by others working on image processing and deep learning solutions for poultry disease detection, as previous work did not use object detection before image analysis, they noted.
This omission could lead to non-target objects in photos affecting the learning process, while the new models have been successfully adapted into an accessible format for those likely to use it — an essential factor in the success of ‘Agriculture 4.0’ technologies.
“Our system has a capability to identify most common poultry diseases from chicken faecal images. The developed system can be used in poultry farms to assist farmers and veterinarians. It can be also used as baseline for other researchers for further improvement,” they added.
“More dataset collection, especially for diseases that are not included in this study can improve the accuracy and quality of the system.”
