In agriculture, it is really difficult to precisely distinguish the diseases, the pests that have infected your crop, and also some nutrient deficiencies, because of the symptoms’ similarities, especially at the beginning of their emergence. This task, although it seems to be easy even for farmers and agronomists, has the possibility to cause serious problems to crop production, due to the potential lack of correct detection and improper treatment efforts. Computer vision assists the effective control of disease and pest infestation through image collection from the field and annotation of the infected areas of the crops by domain experts with accuracy and validation.
As you can see in the next photo there are 2 orange tree nutrient deficiencies, having almost the same symptoms. From the agronomic point of view, the sooner a nutrient deficiency is detected, the less impact it will have on crop production because the farmers’ reaction will be to fertilize the orchard, or only the plants having the deficiencies and boost the trees. Although it is common practice for farmers to analyze the leaves for the nutrient content in specialized chemical laboratories, this method is rather slow and nonrepresentative for each and every plant, but the orchard in total.
Accordingly, when it comes to pest infestations and disease damages, any early control latency may lead to reduced production and loss of producer revenue. In the following image, there is a tomato plant infected by the pests Tuta absoluta and Liriomyza. The purple bounding boxes represent Tuta absoluta and the green ones are Liriomyza. Both of them cause severe damage to the tomato plants but especially Tuta absoluta, is the most dangerous tomato plant pest enemy that if not treated immediately, can destroy the entire production within a few days. Therefore, it is vital to be detected just when the first symptom occurs. And what if the field has thousands of plants? Is this an easy task for the farmer to make everyday checks on all the plants? The answer is obviously no. Thus, machinery with computer vision equipment is essential to do this task for the farmer.
As a result, from the computer vision point of view, a detection model is able to assist the farmers and the agronomists in decision-making without much labor and time spent. Images have been collected through various time periods and hours of the day in many steps of the infestations or deficiencies and then are annotated by domain experts like entomologists and phytopathologists. The annotated images are then used to train machine learning algorithms and models capable of detecting the disorders. In other words, this practice is able to give added value and escalate crop production in terms of economic prosperity, crop protection, and environmental sustainability.