Edge Detection of Strawberries Ripeness Based on Model Optimization Using Intel OpenVINO Toolkit

Abstract

The improvement of automation and big data analytics with technology is giving big benefits to the agriculture sector.In the open field crops, farming robot technology helps farmers to spray fertilizer.Soil data analysis can help to determine plant treatment.Harvesting robots for picking fruits with computer vision and robotics can support the productivity and quality of crops.To implement this, appropriate hardware and algorithms are important to define.One of the most premium fruits that can be cultivated by using high technology is strawberries.We need to declare the best trade-off between system design (software & hardware) with implementation especially in agricultural sector.In this experiment, the YOLOX algorithm is running to detect the ripeness of strawberries.The algorithms run in two modes: GPU and CPU only.The best results show that the YOLOX-S algorithm, which runs in GPU mode, is 95.75% in precision and 59 fps in throughput.It will be difficult to accommodate a harvesting robot processor that has a GPU.The algorithm is now run in CPU only mode and it gives only 11.31 fps in throughput.Then the proposed model, which is already optimized by Intel OpenVINO, gives better results in throughput, showing 37.15 fps.So, with the proposed optimized model, we can choose CPU-only hardware for affordable hardware implementation.