The report covers a period from 1st September 2016 to 31st August 2017. During the reporting period work was commenced in all areas of work: Hardware and Software development, Development of Image recognition and Machine Learning algorithms, Trap network building, Sales, Communication and Dissemination, etc.
The most notable results have been achieved in the area of HW development and Image Recognition. We have successfully launched trap self-cleaning mechanism, which reduces the need for manual changing of the sticky plates and thus reducing growers’ travel expenses. The innovation used in the network of already saved thousands of kilometers and hours of manual labor, proved to be reliable and is fully capable of reaching the objective set out in the project charter. As experienced in one of the cases, one self-cleaning mechanism saved the grower 17(!) travels to the trap in only one month.
One of the main goals of the project is to develop and improve accuracy of the state-of-the-art computer vision system. Initial results of testing of newly developed Deep Learning Neural Networks algorithm have significantly surpassed our expectations as the new algorithm is achieving more than 90% of accuracy. It is necessary to point out that such accuracy is far better than human and anything else that is currently available in the market. Nevertheless, there is plenty of work to be done especially in the area of continuous learning process of neural networks and increasing robustness of the algorithms when it comes to mixed pest population on the same sticky plates (avoiding false positives).
By the date of this report there were eleven trap clusters in operation consisting of several hundred traps in 11 South European countries. The network is made of the same type of traps, attracting insects with the same pheromone lure and thus enabling statistical consistency of the data collected. Unification of the traps and the number of traps in the network itself provide consistent quality data, which is also statistically representative. This is vital for building statistical models and data analysis in order to deliver pest forecast.
The project also allowed us to reach out to much larger audience by attending large number of different events (pitching, investors’, etc.), tradeshows, fairs, specialized events for growers and be present in the general public media (national television and newspapers). All the activities in the communication and dissemination area reflected also in the sales results as we have been able to successfully follow the company growth plans outlined in the project charter.
