Accurate Diagnosis of Tomato Leaf Diseases Enabled by AI Solution

Plant diseases have posed a major threat to farmers since the early days of agriculture. Today, despite our improved understanding of the causes and treatment of these diseases, they still cause significant economic losses. Although detecting plant diseases early is a grower’s best bet to minimize their impact, manually checking each plant is a huge task and prone to errors. Only a well-trained eye can accurately distinguish between diseases that cause similar symptoms.

Fortunately, artificial intelligence (AI) is quickly paving the way for smarter farming practices. State-of-the-art machine learning models are able to automatically identify plant diseases from digital images. When combined with high-quality drones and cameras, these models can reduce the time and effort required to monitor large fields. However, even the newest algorithms struggle under specific difficult conditions.

One notable example is the effect of background interference on disease classification outcomes. In some cases, the diseased leaves acquire a color similar to that of the soil, confusing the automatic classifier, especially when the affected areas are on the leaf edges. Other problems include the variety of symptoms caused by a single disease and the similarities that exist between different diseases.

In a new study, a team of researchers set out to develop a model that could address these challenges. They focused on five common diseases that affect tomato leaves and developed a machine learning model called PLPNet, which can accurately detect these diseases from images captured in real time. The study, which was led by Professor Guxiong Zhu of the Central South China University of Forestry and Technology, was recently published in Plant Phenomenology .

The team first focused on producing a good dataset to train the model. To this end, they collected images from an open but rather old dataset called Botanical Village. They analyzed the images thoroughly and removed those that didn’t make them good candidates for training, such as faded or insufficiently lit photos. In addition to the final 3,524 images they obtained from Plant Village, the team also downloaded another 1,909 images from the Internet. Finally, careful classification of all images was performed to identify each lesion on the leaves.

Next, the team designed the PLPNet network architecture. They used three distinct techniques that, working together, resulted in the highest classification accuracy. The first was the backbone of perceptual adaptive convolution (PAC), which helped the model extract the most specific characteristics of each disease by adjusting the ‘focus’ of the network when analyzing an image.

The second was the Location Reinforcement Attention Mechanism (LRAM) module, which helped detect diseases on leaf edges and filter out background interference. The third module was a proximity feature aggregation network (PFAN) implementing switchable and disassembled atrial torsion. This structure helped the model know the smallest details of each disease, which greatly improved its performance in disease detection and classification.

The team thoroughly tested their model after training and analyzed the performance gained for each of its parts. They also compared the performance of PLPNet to several other recent models for detecting plant diseases.

The results were very promising, with PLPNet achieving 94.5% accuracy at over 25 frames per second, making it suitable for field use. Enthusiastic about the results, Professor Chu notes, “PLPNet significantly improves detection accuracy while maintaining standard detection speed. Thus, it is superior to other test paradigms and demonstrates the effectiveness of our enhanced approach.”

Tomatoes are widely cultivated all over the world and are of great economic importance. The team expects PLPNet to have a positive impact on their cultivation, reducing the burden of financial losses caused by diseased tomato plants. “This research can help producers detect tomato leaf diseases in a timely and accurate manner, as well as set specific controls based on the type of disease detected,” concludes Professor Zhu. “This provides a new reference for deep learning in ensuring modern tomato cultivation.”