differential spraying in precision agriculture

One of the objectives of precision agriculture is to reduce the use of herbicides by using site-specific weeds detection by computer vision. Weed detection and the decision whether to spray or not herbicides in small fields is not a major challenge. However, this decision becomes hard when dealing huge fields of hundreds of hectares needs to be sprayed. Overuse of herbicides produces ecological footprints with negative long term effect on the soil. It further poses health hazards, and the heavy expenses on the farmer’s side. Homogenous spraying is not economic, especially due to the observation that weed grows in patches rather than in homogenous distribution.
It therefore, is important to detect and differentially spray specific sites of the field where patches of weed is found. In terms of quantification, the total proportion of weeds in the field is important quantity because it indicated the extent of competition between weeds and crop. This quantity is especially important during early stages of the crop growth, in which the weed aggressively competes with crops on resources and space.
Several automatic procedures based on computer vision to detect weed in fields has been proposed in the past. Most of them face the challenge of differentiating the weed from crops by means of geometrical information such as shape and texture, as well as the weed’s distribution within the field. Isolated weeds are easy to detect automatically, especially if they grow off the grid. However, the real challenge is to detect those weeds which grow in-row within the crops. But detection is only the first step in differential spraying, the second step is to make the decision whether to spray or not.
In a paper published in the journal of Computers and Electronic in Agriculture, Alberto Tellaeche et al sought out to devise an automatic computer-vision based system for detecting a weed called. Their framework is composed of two sub-systems, the first is image segmentation, and the second is the decision making engine. The image processing sub-system consists of an initial stage of image binarization by a threshold using the entropy of the histogram. Morphological operations then followed to clean the binary image from spurious points. These operations resulted in a binary image, for which the authors report mostly crop and weed are present. The rows of the plantation were then detected by a Hough transform, and the image was partitioned into cells. Analysis then followed in each one of the cells in the region close to the camera viewpoint. Based on the output image, a multi-criteria fuzzy decision making subsystem kicked into action. In essence this system has to make the choice whether or not to spray a cell, making it a binary classifier.
The system was put to the test using a set of 146 images, half of which were acquired on a sunny day and the other half on a cloudy day. The authors randomly split the images into 3 subsets, and applied their method. The result of the fuzzy classifier was tested against human expert, who inspected and detected the weed in the images. A total of 5 tests were performed for which the system’s correct classification percentage ranged from 73 to 92%. Thus, demonstrating the usability of the proposed method in automatic weeds detection by computer vision and management of differential spraying of herbicides.

Reference
Tellaeche, Alberto, et al. “A new vision-based approach to differential spraying in precision agriculture.” computers and electronics in agriculture 60.2 (2008): 144-155.

Add a Comment

Your email address will not be published. Required fields are marked *

Prove you're a human *